JP2005031789A - Object type shape discrimination method and system using RBF network using image information - Google Patents

Abstract

An object of the present invention is to provide an object type shape discrimination method using an RBF network that can discriminate the types of various objects and uses image information with a wide range of applications.
An object type shape discrimination method using an RBF network using image information according to the present invention includes a first step of obtaining image information of an object, and extracting a predetermined feature parameter relating to the image from the image information of the object. The second step 22 and the predetermined feature parameter relating to the image obtained in the second step 22 are taken into the RBF network, the probability of belonging to each object type is calculated by the RBF network, and the calculation result is assigned to the predetermined class. The third step 23 is capable of discriminating the types of various objects from the probability of belonging.
[Selection] Figure 4

Description

【００４７】
なお、数式１では、分子の「画像データの一次微分（ｐｒｅｗｉｔｔオペレータ）」は輪郭の強さと方向を表すため、船舶の画像データ中の船舶の輪郭を中心に強調させるためのものである。また、分母の「画像データの二次微分（ｌａｐｌａｃｉｓｎオペレータ）は、輪郭の強さのみを表すため、船舶の画像データ中の船舶の輪郭以外に背景や船舶像の中の点状のノイズ成分も同時に強調するためのものである。
【００４８】
ついで、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａで実行されている遺伝子処理ステップ２２１は、選択、淘汰（この実施の形態では、上位５０％存続：個体数２５個）の遺伝的処理を実行する（Ｓ２２１３）。
次に、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａで実行されている遺伝子処理ステップ２２１は、交差（この実施の形態では、二点交差法：個体数５０個）といった遺伝的処理を実行する（Ｓ２２１４）。
さらに、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａで実行されている遺伝子処理ステップ２２１は、突然変異（この実施の形態では、確立０．１［％］といった遺伝的処理を実行する（Ｓ２２１５）。
【００４９】
また、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａで実行されている遺伝子処理ステップ２２１は、前述した遺伝的処理を施した後のデータを基に、再び数式１による適応度を求める（Ｓ２２１６）。
また、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａで実行されている遺伝子処理ステップ２２１は、適応度が所定の世代（この実施の形態では、例えば１０世代（図６参照））に渡って一定値に達していなければ（Ｓ２２１７；ＮＯ、）、ステップ２２１３に戻って再び上記ステップ２２１３、２２１４、２２１５、２２１６を繰り返し処理する。
【００５０】
なお、上記処理ステップの処理がされることにより、この実施の形態では、図６に示すように、実線は個体数３０個の場合の１０世代の適応度の変化を示し、点線は個体数５０個の場合の１０世代の適応度の変化を示し、一点鎖線は個体数７０個の場合の１０世代の適応度の変化を示している。
一方、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａで実行されている遺伝子処理ステップ２２１は、選択、淘汰、交差、突然変異といった遺伝的処理を行い（Ｓ２２１３〜２２１５）、再び数式１による適応度を求める操作を行い（Ｓ２２１６）、図６に示すように、適応度が所定の世代（この実施の形態では、例えば１０世代）に渡って一定値に達したときに（Ｓ２２１７；ＹＥＳ）、最も適応度の高い遺伝子配列が得られたことになる。
【００５１】
前記コンピュータ装置３の制御処理部３１のＣＰＵ３１ａで実行されている遺伝子処理ステップ２２１は、前記ステップにおいて得られた最大適応度による遺伝子配列に基づく二値化をする（Ｓ２２１８）。上述したようにコンピュータ装置３の制御処理部３１のＣＰＵ３１ａで実行されている遺伝子処理ステップ２２１により処理された結果、前述したように例えば図１２（ｂ）に示すように自動車運搬船が鮮明に表示できるし、同様に、例えば図１３（ｂ）に示すようにタンカーが鮮明に表示できるようになる。
【００５２】
このような遺伝子処理ステップ２２１が終了すると、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａは、基準線算出処理ステップ２２２に移行する。前記制御処理部３１は、基準線算出処理ステップ２２２に処理を開始すると、例えば図１４に示すタンカーの画像データに対して、例えば図１５に示す処理を実行して、前記遺伝子処理ステップ２２１で抽出した船舶画像データに対して上甲板線を求める処理を行う。
【００５３】
すなわち、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａで処理される基準線算出処理ステップ２２２では、図１４に示すタンカーの画像データ３７０における上甲板線（基準線）ＳＬを求めることになる。この上甲板線の検出を行う理由は、船体と上部構造物を分離するために必要だからである。
このＣＰＵ３１ａで処理されている基準線算出処理ステップ２２２では、通常、船舶像を水平方向に投射した投射の最大値を示す位置を上甲板線（基準線）ＳＬとする処理をおこなう。
【００５４】
しかし、上甲板線ＳＬが、前述した位置よりも下に存在する船種に対しては、前記基準線算出処理ステップ２２２は、図５（ａ）及び図５（ａ’）に示すように、水平方向に投射した射影の最大値を示す位置ＭＰから喫水位置ＷＰの範囲で、ｐｒｅｗｉｔｔオペレータによりエッジを検出し（図１５（ａ）及び図１５（ａ’）のＳ２６１）、これによりエッジ画像３８０ｂを得て、その得たエッジ画像３８０ｂに対して細線化処理を施し（図１５（ａ）及び図１５（ａ’）のＳ２６２）、これにより細線化画像３８０ｃを得て、さらに、前記細線化画像３８０ｃに対してＨｏｕｈ変換を施すことにより求めた画像３８０ｄに示す直線部分を上甲板線ＳＬとした（図１５（ａ）及び図１５（ａ’）のＳ２６３）。
【００５５】
このように前記基準線算出処理ステップ２２２において求めた上甲板線（基準線）ＳＬを基に、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａによって実行されている強調処理ステップ２２３は、上甲板線上の構造物を強調する。この基準線算出処理ステップ２２２によって処理された結果、図１６（ａ）に示すコンテナ船の画像データ３８０ａについては、図１６（ｂ）に示すようにコンテナ船に引かれた上甲板線（基準線）ＳＬ線の上側が強調された画像データ３８０ｂになる。
【００５６】
同様に、図１７（ａ）に示す自動車運搬船の画像データ３９０ａについては、図１７（ｂ）に示すように自動車運搬船に引かれた上甲板線（基準線）ＳＬ線の上側が強調された画像データ３９０ｂになる。同様に、図１８（ａ）に示すばら積み船の画像データ４００ａについては、図１８（ｂ）に示すようにばら積み船に引かれた上甲板線（基準線）ＳＬ線の上側が強調された画像データ４００ｂになる。同様に、図１９（ａ）に示すタンカーの画像データ４１０ａについては、図１９（ｂ）に示すようにタンカーに引かれた上甲板線（基準線）ＳＬ線の上側が強調された画像データ４１０ｂになる。
【００５７】
ついで、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａは、次のステップである特徴パラメータ処理ステップ２２４の処理に移行する。制御処理部３１のＣＰＵ３１ａで処理されている特徴パラメータ処理ステップ２２４は、前記船舶全体画像データと、前記強調処理ステップ２２３で求めた上部構造物の画像データとを基に、船舶全体と上部構造物の面積比Ｒ、円形度Ｃ及び図形重心の高さＨなどの特徴パラメータを求める。
【００５８】
上記強調処理ステップ２２３で算出する特徴パラメータとは、船舶全体と上部構造物の面積比Ｒ、円形度Ｃ、図形重心の高さＨなどである。
まず、特徴パラメータの一つである船舶全体と上部構造物の面積比Ｒは、
Ｒ＝Ｓｕ／Ｓａ …［２］
Ｓｕ：上部構造物の面積
Ｓａ：船舶全体の面積
で与えられる。
また、特徴パラメータの他の一つである円形度Ｃは、
Ｃ＝４πＳｃ／Ｌ^２ …［３］
Ｓｃ：重み付け後の船舶像の面積
Ｌ：重み付け後の船舶像の輪郭長
で与えられる。
【００５９】
また、特徴パラメータのさらに他の一つである図形重心の高さＨは、
Ｈ＝（２ＲＹ／Ｗ）ｔａｎ（φ／２） …［４］
Ｒ：相手船までの距離
Ｙ：重み付け後の垂直方向の図形重心
Ｗ：画像の垂直方向の全画素数
φ：カメラの角度
で与えられる。
【００６０】
また、特徴パラメータのさらに他の一つである水平方向の図形重心Ｄは、
Ｄ＝Ｘ／Ｙ
Ｘ：重み付け後の水平方向の図形重心
Ｙ：船舶の水平方向の画素数
で与えられる。
このようにして特徴パラメータ処理ステップ２２４で求めた特徴パラメータを用い、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａにおいて第３のステップ２３を処理することにより、次のようにして物体（船舶）の種別や形状を判定する。
【００６１】
前記コンピュータ装置３の制御処理部３１のＣＰＵ３１ａにおいて処理されてる第３のステップ２３は、図２２に示すＲＢＦ（ＲａｄｉａｌＢａｓｉｓＦｕｎｃｔｉｏｎ）ネットワーク５０を使用し、前記船舶全体画像データと前記強調処理ステップ２２３で求めた上部構造物の画像データとを基に、船舶全体と上部構造物の面積比Ｒ、円形度Ｃ及び図形重心の高さＨの特徴パラメータを入力要素とし、当該入力要素をＲＢＦネットワークに取込み、当該ＲＢＦネットワーク５０によって各物体種別に属する確率を演算し当該演算結果が所定のクラスに属する確率から各種物体の種別（船舶の種別）を分類している。
【００６２】
なお、ＲＢＦネットワーク５０は、通常、上述した属性から各種物体の種別（船舶の種別）を判別することができるが、例えば図２０に示すようにコンテナ船と、漁船とが同一特徴範囲に分類されたときには、さらに、別の入力要素を基にして確率を計算することにより、図２１に示すように、タンカーと、漁船とを分類することができる。
【００６３】
このような処理ステップを説明するために、ＲＢＦネットワーク５０は、図２２に示すように、第１分類５２において所定の特徴パラメータ（Ｒ、Ｃ、Ｈ）を用いて物体の種別（船種）を分類し、第１分類５２で同一範疇に分類されたもの（タンカー、漁船）を、さらに別な特徴パラメータ（Ｈ、Ｄ）を用いて第２分類５４で同一分類る分類されたものをさらに他の物体（船種（タンカーと、漁船とに分類）として分類している。
しかし、上記ＲＢＦネットワーク５０は、当該ネットワーク５０による確率計算で物体の形状が判別できるときには、この第１分類５２と、第２の分類５４とは必須の要件ではない。
【００６４】
なお、上述した船舶全体と上部構造物の面積比Ｒと、円形度Ｃと、図形重心の高さＨを特徴パラメータを第１主成分と第２主成分に分けて５種類の船舶をプロットしたものが主成分得点の散布図（図２０）である。この図２０からもわかるようにタンカーと漁船は、得点分布の一部が重複していることがわかる。
そこで、第一段階での船種の判別では両者を同一カテゴリーとして扱うことにして処理を進めることとして説明をしてゆく。
【００６５】
次に、この同一カテゴリー内の分類を行うために、図形の重心の高さＨと水平方向の図形重心Ｄの２つの特徴パラメータを用いて主成分分析を行った結果が図２１に示す散布図である。この図２１の散布図からもわかるようにタンカーと漁船が明らかに分類されていることがわかる。
この遺伝的アルゴリズムによる船種判別システム１では、コンピュータ装置３で実行されている第３のステップ２３においてＲＢＦネットワーク５０を用い、ＲＢＦネットワーク５０による確率計算によって、物体の種別（船種）の判別を行っている。
【００６６】
なお、本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システム１では、ＲＢＦネットワーク５０は、前述した主成分分析の結果を基に作成し、図２２に示すように、第一分類５２において船舶全体と上部構造物の面積比Ｒ、円形度Ｃ、図形重心の高さＨの３つの特徴パラメータを入力要素として、「自動車運搬船」、「コンテナ船」、「バラ積み船」の他、「タンカー」と「漁船」を同一カテゴリーとして判別をおこない、第二分類５４において図形重心の高さＨと水平方向の図形重心Ｄの特徴パラメータを入力要素として「タンカー」と「漁船」の判別を行う二段階構成としてある。また、本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムでは、学習はアスペクト１０度ごとの船舶モデルの船舶像から算出した特徴パラメータを用いて行っている。
【００６７】
さらに、本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システム１では、船舶判別は上述した数式２〜数式５に示すように、相手船までの距離、アスペクトなど、それぞれ条件の異なる船舶の赤外画像を用いて行っている。
このように画像情報を利用したＲＢＦネットワークによる物体種別形状判別システム１において、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａが画像情報を利用したＲＢＦネットワークによる物体種別形状判別処理プログラム２０を処理した結果、表１のようなデータが得られた。
【００６８】
【表１】

【００６９】
この表１に示すように、自動車運搬船については試験画像数１１枚に対して１００［％］の判別率であり、コンテナ船については試験画像４枚に対して７５［％］の判別率であり、ばら積み船については試験画像数６枚に対して１００［％］の判別率であり、タンカーについては試験画像１３枚に対して１００［％］の判別率であり、漁船については試験画像５枚に対して８０［％］の判別率であった。
【００７０】
なお、画像情報を利用したＲＢＦネットワークによる物体種別形状判別システム１において、コンピュータ装置３の制御処理部３１のＣＰＵ３１ａで処理した結果は、ハードディスク装置３３のデータベース領域１６に格納しておく。
本発明に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別方法及びそのシステム１では、ほぼ全ての船種において、良好な結果を得ることができた。また、誤判別を生じた船舶について検討した結果、船舶の判別精度は２値化の処理結果に大きく依存することがわかった。
【００７１】
以上説明したように本発明に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別方法及びそのシステム１は、ＲＢＦネットワーク５０を用いた物体種別（船種）判別により、ほぼ全ての物体の種別（船種）において、良好な結果を得ることができ、かつ運動性能の収集もできる。このような物体の種別を判別することが可能であれば、例えば船舶の安全運行を支援する重要な要素である船種を確実に判別することができるという利点がある。
【００７２】
以上説明したように本発明では、画像情報を利用したＲＢＦネットワークによる物体種別形状判別方法及びそのシステム１の好適な実施の形態として、物体として船舶を例に挙げて説明したが、この物体としては船舶に限定されるものではなく、本発明に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別方法及びそのシステムによれば、自動車、戦車、人、その他、あらゆる物体の種別や形状を判別することがかものであることはいうまでもない。
また、本発明に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別方法及びそのシステムでは、物体の画像として赤外線画像を使用したが、これに限定されることなく、物体を通常の昼光色で撮像した画像であっても、その他の光による物体を撮像手段によって撮像した画像であってもよい。
【００７３】
【発明の効果】
以上説明したように本発明に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別方法及びそのシステムによれば、遺伝的アルゴリズムにより物体像を抽出し、その抽出した物体像を所定の基準線を基に構造物の強調をし、その強調した画像から特徴パラメータを求め、その特徴パラメータをＲＢＦネットワークに入力して物体の種別や形状の判別を行うことができる。
また、本発明に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別方法及びそのシステムでは、物体の種別を判別した結果をいろいろな技術分野（例えば、衝突回避システムなど）に応用することができるという利点がある。
【図面の簡単な説明】
【図１】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムを示すブロック図である。
【図２】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムの一部を構成するコンピュータ装置を説明するためのブロック図である。
【図３】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおけるハードディスク装置に格納されているプログラムやデータを説明するための図である。
【図４】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムによる画像情報を利用したＲＢＦネットワークによる物体種別形状判別処理プログラムの構成例を示す図である。
【図５】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおけるコンピュータ装置で処理する遺伝子処理ステップの具体的な処理フローチャートである。
【図６】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおけるコンピュータ装置で処理する遺伝子処理ステップで適応度が変化してゆく様子を示す特性図である。
【図７】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、赤外線デジタルカメラからコンピュータ装置に取り込んだ自動車運搬船の画像データの例を示す図である。
【図８】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、赤外線デジタルカメラからコンピュータ装置に取り込んだタンカーの画像データの例を示す図である。
【図９】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、船舶画像データを横方向の遺伝子配列と縦方向に遺伝子配列に区分けした図である。
【図１０】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、遺伝子処理ステップにおける最大適応度を算出した遺伝子配列に基づく分割領域をコンテナ船の赤外画像データに適用し、かつ遺伝子処理した後の結果の抽出画像データを示す図である。
【図１１】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、遺伝子処理ステップにおける最大適応度を算出した遺伝子配列に基づく分割領域をタンカーの赤外画像データに適用し、かつ遺伝子処理した後の結果の抽出画像データを示す図である。
【図１２】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、赤外線画像データを遺伝子処理ステップで処理した結果の画像データを示す図である。
【図１３】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、赤外線画像データを遺伝子処理ステップで処理した結果の画像データを示す図である。
【図１４】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、基準線算出処理ステップにおける基準線について説明するための図である。
【図１５】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、基準線算出処理ステップの動作を説明するための説明図である。
【図１６】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、前記遺伝子処理ステップで得た画像データを、構造物強調処理ステップによって強調した場合の画像データを示す図である。
【図１７】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、前記遺伝子処理ステップで得た画像データを、構造物強調処理ステップによって強調した場合の画像データを示す図である。
【図１８】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、前記遺伝子処理ステップで得た画像データを、構造物強調処理ステップによって強調した場合の画像データを示す図である。
【図１９】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、前記遺伝子処理ステップで得た画像データを、構造物強調処理ステップによって強調した場合の画像データを示す図である。
【図２０】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、５種類の船舶を第１主成分と第２主成分にプロットした図である。
【図２１】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて、２種類の船舶を第１主成分と第２主成分にプロットした図である。
【図２２】本発明の実施の形態に係る画像情報を利用したＲＢＦネットワークによる物体種別形状判別システムにおいて使用するＲＢＦ（ＲａｄｉａｌＢａｓｉｓＦｕｎｃｔｉｏｎ）ネットワークを示す説明図である。
【符号の説明】
１ 遺伝的アルゴリズムによる船種判別システム
３ コンピュータ装置
５ キーボード（入力装置）
７ マウス（入力装置）
９ ディスプレイ（出力装置）
１０ ＯＳ格納領域
１１ プリンタ（出力装置）
１２ ＡＰ格納領域
１３ 赤外線デジタルカメラ（撮像手段）
１４，１６ データベース領域
２２ 画像取得処理ステップ
２４ 船舶画像データ抽出処理ステップ
２６ 基準線算出処理ステップ
２８ 構造物強調処理ステップ
３０ 船種判別処理ステップ
３１ 制御処理部
３１ａ ＣＰＵ
３１ｂ ＲＯＭ
３１ｃ ＲＡＭ
３２ 主メモリ
３３ ハードディスク装置
３４ 入力用Ｉ／Ｆ
３５ 表示用Ｉ／Ｆ
３６ プリンタＩ／Ｆ
３７ ＵＳＢボード
３８ バスライン[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an object type shape discriminating method by an RBF network using image information where the type and shape of various objects can be discriminated from image information of the object, and a system thereof.
[0002]
[Prior art]
In general, if the shape and type of an object can be determined from image information of the object captured by an imaging device such as an electronic camera or a TV camera, it can be easily applied to various technical fields.
For example, in the case of a collision avoidance system that supports safe operation of a ship, in addition to knowing the course and movement performance of the other ship, if the shape and type of the other ship can be determined, It is possible to avoid collision effectively.
The collision avoidance system having such a system configuration can be said to be an extremely effective system in a situation where the risk of a collision accident between ships is increasing due to an increase in the amount of ship traffic as in today's present day.
By the way, the present inventors have proposed a ship collision avoidance system based on a genetic algorithm as a ship collision avoidance system based on the course and movement performance of the own ship and the course and movement performance of the other ship (for example, Patent Document 1).
[0003]
In this conventional collision avoidance system using a genetic algorithm, the fitness A, which is an evaluation of the moving path of a moving object, is determined by the safety S, the deviation D from the target path, the economic efficiency E, and the immediate death determining gene B. = B × (S + D + E), and the numerical value of safety S is logarithmically smaller with respect to the narrowing of the interval between moving objects that move relatively.
Japanese Patent Laid-Open No. 2002-222500.
[0004]
[Problems to be solved by the invention]
However, the above-described conventional collision avoidance system of a genetic algorithm seeks the possibility of a collision based on the course and movement performance of the own ship and the course and movement performance of the other ship, and provides safe operation. It did not identify the ship type, which is an important element to support.
SUMMARY OF THE INVENTION An object of the present invention is to provide an object type shape discrimination method and system using an RBF network that solves the above-described problems, makes it possible to discriminate the types of various objects, and uses image information with a wide application range.
[0005]
[Means for Solving the Problems]
In order to achieve the above object, an object type shape discrimination method by an RBF network using image information according to claim 1 is an object type discrimination method capable of discriminating an object type from image information of an object. A first step of obtaining the image information, a second step of extracting a predetermined feature parameter relating to the image from the image information of the object, and a predetermined feature parameter relating to the image obtained in the second step to the RBF network And a third step of calculating the probability of belonging to each object type by the RBF network and determining the type of various objects from the probability that the calculation result belongs to a predetermined class.
According to a second aspect of the present invention, in the object type shape discrimination method by the RBF network using the image information according to the first aspect, the second step is based on a genetic sequence for the image information of the object. The image information is segmented into image data, and the genetic sequence having the highest fitness is obtained while exploring genetic processing on the image data to obtain segmented image data, and for each segment of the segmented image data. A gene processing step for extracting object binarized image data by discriminating and analyzing the data, and a reference line for obtaining a reference line for the object image with respect to the object binarized image data extracted in the gene processing step A calculation processing step, and an enhancement processing step for obtaining emphasized image data by emphasizing a predetermined structure based on the reference line in the object binarized image data. And flop, characterized in that a characteristic parameter processing step for calculating a feature parameter concerning the image on the basis of the weighted image data obtained by the image information and the enhancement processing step of said object.
According to a third aspect of the present invention, in the object type shape discrimination method by the RBF network using the image information according to the first aspect, the first step obtains image information of the object by imaging the object with an imaging means. It is characterized by that.
According to a fourth aspect of the present invention, in the object type shape discrimination method by the RBF network using the image information according to the first aspect, the third step includes a feature parameter relating to an image obtained from the image information of the object and the image. An RFB network that calculates a probability that a feature parameter related to an input image belongs to a predetermined class based on a difference between learning data of feature parameters related to the parameter and a parameter that determines an allowable range is used. A feature parameter is input, and the probability is calculated by the RFB network to determine the type of the object.
In order to achieve the above object, an object type shape determination system using an RBF network using image information according to claim 5 is an object type determination system capable of determining an object type from image information of an object. Imaging means for obtaining the image information, feature extraction means for extracting a predetermined feature parameter for the image from the image information of the object from the imaging means, and a predetermined feature parameter for the image obtained by the feature extraction means And an object type discriminating unit which can take in the network, calculate the probability of belonging to each object type by the RBF network, and discriminate the type of various objects from the probability that the calculation result belongs to a predetermined class.
According to a sixth aspect of the invention, in the object type shape discriminating system using the RBF network using the image information according to the fifth aspect, the feature extracting unit is configured to perform the processing based on a genetic sequence with respect to the image information of the object. The image information is segmented into image data, and the genetic sequence having the highest fitness is obtained while exploring genetic processing on the image data to obtain segmented image data, and for each segment of the segmented image data. Gene processing means for extracting object binarized image data by discriminating analysis and binarization, and a reference line for obtaining a reference line for the object image for the object binarized image data extracted by the gene processing means Calculation processing means, enhancement processing means for obtaining emphasized image data by emphasizing a predetermined structure based on the reference line in the object binarized image data, and Characterized in that a characteristic parameter processing means for calculating a feature parameter relating to image the enhanced image data obtained by the image information and the enhancement processing unit body based.
According to a seventh aspect of the invention, in the object type shape discriminating system using the RBF network using the image information according to the fifth aspect, the object type discriminating means includes a feature parameter relating to an image obtained from the image information of the object and the image. And an RFB network that calculates a probability that a feature parameter related to an input image belongs to a predetermined class based on a difference between learning data of feature parameters related to the parameter and a parameter that determines an allowable range. A feature parameter is input, and the probability is calculated by the RFB network to determine the type of the object.
[0006]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
1 to 22 are diagrams for explaining an object type shape discrimination method using an RBF network using image information and an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention. In these drawings, the object type shape discrimination method by the RBF network using the image information according to the embodiment of the present invention can be applied to a ship collision avoidance system. Hereinafter, it is applied to a ship collision avoidance system. An example will be described.
[0007]
In the present invention, in order to discriminate the type (ship type) of an object, an object image (ship image) for each aspect of an object to be discriminated (a ship in this embodiment) is required. Here, the aspect is an angle formed by the direction of the opponent's object (ship) and the course of the opponent's object (ship) as viewed from yourself (your ship). However, it is difficult to obtain object images (ship images) of all aspects for all objects (ships).
[0008]
Therefore, in the present embodiment, a predetermined object image is substituted. In the present embodiment, as a predetermined object image, a ship having a standard hull form among ship types that occupy about 69% of collision accidents (“container ship”, “tanker”, “bulk ship”, “fishing ship”). , “Automobile carrier”), a ship model is created by computer graphics, feature parameters for each aspect 10 [degree] are calculated from these ship images, and then, the calculated feature parameters are affected by the aspect. In order to select a ship type that can be classified well, an analysis using an RBF network is performed for each aspect 10 [degree], and the type of the object (ship type) is determined. This will be specifically described below.
[0009]
FIG. 1 is a block diagram showing an object type shape discrimination system by an RBF network using image information according to an embodiment of the present invention. In FIG. 1, an object type shape discrimination system 1 by an RBF network using image information according to an embodiment of the present invention is roughly divided into a computer device 3, a keyboard 5 and a mouse 7 as input devices, and an output. The apparatus includes a display 9 and a printer 11 as apparatuses, and an infrared digital camera 13 as an imaging unit. A keyboard 5, a mouse 7, a display 9, a printer 11, and an infrared digital camera 13 are connected to the computer device 3.
[0010]
The infrared digital camera 13 is not always connected to the computer apparatus 3 but is connected to the computer apparatus 3 by, for example, USB connection when an image to be processed is taken into the computer apparatus 3. It is a thing. In the above embodiment, the infrared digital camera 13 is used as the imaging means. However, this is only adopted because it can capture images even at night, and only image information captured during the daytime may be used. Needless to say, an ordinary digital camera may be used.
Further, the ship image captured by the infrared digital camera 13 is obtained by performing a binarization process described later, calculating a feature parameter, and determining a ship type by the RBF network.
[0011]
FIG. 2 is a block diagram for explaining a computer device that constitutes a part of an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
The computer 3 includes a control processing unit 31 that executes object type shape determination processing by the RBF network using various image information and other processing and control, and object type shape determination by the RBF network using the operating system and image information. A main memory 32 for storing processing programs and other necessary data, and a hard disk for storing an object type shape discrimination processing program, a ship image database, and other necessary data and databases by an RBF network using the operating system and the image information The apparatus 33 includes an input interface (I / F) 34, a display I / F 35, a printer I / F 36, a USB board 37, and a bus line 38 for connecting them.
[0012]
The control processing unit 31 includes a CPU 31a that executes various arithmetic processes, a ROM 31b that stores a program that causes the CPU 31a to execute a predetermined process at the time of initializing, and a CPU 31a that is necessary when the CPU 31a executes arithmetic processes. The RAM 31c temporarily stores data, and these are connected by a bus line 38.
[0013]
In addition, a keyboard 5 and a mouse 7 are connected to the input I / F 34, whereby various data and commands can be given to the control processing unit 31 of the computer device 3. A display 9 is connected to the display I / F 35, so that the result processed by the control processing unit 31 in the computer device 3 can be displayed on the display 9. A printer 11 is connected to the printer I / F 36, so that data to be printed out of the results processed by the control processing unit 31 of the computer apparatus 3 can be given to the printer 11 for printing. An infrared digital camera 13 is detachably connected to the USB board 37, and image data collected by the infrared digital camera 13 is controlled by the control processing unit 31 of the computer device 3 as necessary. The hard disk device 33 of the device 3 can be stored in a predetermined area.
[0014]
FIG. 3 is a diagram for explaining programs and data stored in the hard disk device in the object type shape discrimination system by the RBF network using the image information according to the embodiment of the present invention. In FIG. 3, the hard disk device 33 includes an OS storage area 10 of an operating system and an object type shape discrimination process by an RBF network using image information that realizes an object type shape discrimination method by an RBF network using image information. A program AP storage area 12 and database areas 14 and 16 are provided.
[0015]
When the computer device 3 is turned on and starts operating, the CPU 31a of the control processing unit 31 operates according to the initial program in the ROM 31b, and the operating system stored in the OS storage area 10 of the hard disk device 33 is stored in the main memory. The CPU 31a of the control processing unit 31 operates the operating system on the predetermined storage area of the main memory 32, and the operating system operates. Further, when this operating system is operated, the object type shape discrimination processing program by the RBF network using the image information registered in the operating system in advance is brought into a startable state.
[0016]
In addition, the CPU 31a of the control processing unit 31 in the computer device 3 stores the program in the AP storage area 12 of the hard disk device 33 by activating the object type shape discrimination processing program by the RBF network using the image information with the mouse 7 or the like. The object type shape discrimination processing program by the RBF network using the image information being stored is stored in a predetermined storage area of the main memory 32, and then the image information stored in the predetermined storage area of the main memory 32 When the CPU 31a of the control processing unit 31 processes the object type shape discrimination processing program using the RBF network using the RBF network, the object type shape discrimination system 1 using the RBF network using the image information is realized.
[0017]
An operation of the object type shape discrimination system 1 by the RBF network using the image information having such a configuration will be described. First, an application program for realizing an object type shape discrimination method using an RBF network using image information according to an embodiment of the present invention will be described with reference to FIG.
FIG. 4 is a diagram showing a configuration example of an object type shape discrimination processing program using an RBF network using image information in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
[0018]
In FIG. 4, the object type shape discrimination processing program 20 by the RBF network using image information includes a first step 21 for obtaining object image information, and an image from the object image information obtained in the first step 21. A second step 22 for extracting a predetermined feature parameter relating to the image, a predetermined feature parameter relating to the image obtained in the second step 22 is taken into the RBF network, and the probability belonging to each object type is calculated by the RBF network And a third step 23 for discriminating the types of various objects from the probability that the calculation result belongs to a predetermined class.
[0019]
Here, in the second step 22, the image information of the object is divided into regions based on the genetic sequence to obtain image data, and the image processing is repeated while repeating the genetic processing. Gene processing step 221 that obtains divided image data by exploring a gene sequence with high fitness and extracts object binarized image data by discriminating and analyzing for each region of the divided image data. A reference line calculation processing step 222 for obtaining a reference line for the object image with respect to the object binarized image data extracted in the gene processing step 221, and a structure in the object binarized image data is represented by the reference line. Structure emphasis processing step 223 that obtains emphasized image data by emphasizing a predetermined basis, and object image information obtained in the first step 21 and the structure emphasis The enhanced image data obtained in physical step and a feature parameter processing step 224. for calculating a feature parameter relating to image based.
[0020]
In addition, the third step 23 is configured to input an image based on a difference between a feature parameter relating to an image obtained from the image information of the object, a difference between learning data of the feature parameter relating to the image, and a parameter for determining an allowable range. Using an RFB network that calculates the probability that a feature parameter belongs to a predetermined class, the feature parameter related to the image is input to the RFB network, and the probability is calculated by the RFB network to determine the type of the object It is.
[0021]
The object type shape discrimination processing program 20 by the RBF network using such image information is processed by the control processing unit 31 of the computer apparatus 3, so that the image processing information of the object is stored in the control processing unit 31 of the computer apparatus 3. An image capturing unit for obtaining the image, a feature extraction unit for extracting a predetermined feature parameter for the image from the image information of the object from the image capturing unit, and a predetermined feature parameter for the image obtained by the feature extraction unit is incorporated into the RBF network. Using image information characterized by comprising object type discrimination means for calculating the probability of belonging to each object type by the RBF network and discriminating the type of various objects from the probability that the calculation result belongs to a predetermined class. An object type shape discrimination system 1 using an RBF network is realized.
[0022]
Here, the feature extraction means in the object type shape discriminating system 1 by the RBF network using image information divides the image information into regions based on a genetic sequence for the image information of the object to obtain image data. In addition to obtaining the divided image data by exploring the gene sequence with the highest fitness while repeating the genetic processing on the image data, and discriminating and binarizing each divided image data region Gene processing means for extracting the object binarized image data by means of, a reference line calculation processing means for obtaining a reference line for the object image with respect to the object binarized image data extracted by the gene processing means, and the object binarization Emphasis processing means for obtaining emphasized image data by emphasizing predetermined structures on the basis of the reference line, image information of the object, and the emphasis processing procedure And a feature parameter processing means for calculating feature parameters in an image based on the enhanced image data determined.
[0023]
Further, the object type discrimination means realized by the control processing unit 31 is a parameter for determining a difference between a feature parameter relating to an image obtained from image information of the object and learning data of the feature parameter relating to the image, and an allowable range. Based on the above, an RFB network that calculates the probability that a feature parameter related to an input image belongs to a predetermined class is input. The feature parameter related to the image is input to the RFB network, and the probability is calculated by the RFB network. The type of the object is discriminated.
[0024]
Next, a specific operation of the object type shape discrimination system 1 by the RBF network using the image information will be described with reference to FIGS. 4 to 22 based on FIGS.
Here, FIG. 5 is a specific processing flowchart of the gene processing step processed by the computer device in the object type shape discrimination system by the RBF network using the image information according to the embodiment of the present invention.
[0025]
FIG. 6 is a characteristic diagram showing how fitness changes in a gene processing step processed by a computer device in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention. The horizontal axis represents the generation, and the vertical axis represents the fitness.
FIG. 7 is a diagram showing an example of image data of an automobile carrier ship taken into the computer apparatus from the infrared digital camera in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention. In FIG. 7, the distance between the ship and the infrared digital camera 13 is 2004 [m], the angle of view is 8.2 × 6.2 [degrees], and the sensing element is an SOI diode. In FIG. 7, the shadow of the car carrier is shown by a solid line, but it is unclearly displayed in a specific photographic image.
[0026]
FIG. 8 is a diagram showing an example of tanker image data taken into the computer apparatus from the infrared digital camera in the object type shape discrimination system by the RBF network using the image information according to the embodiment of the present invention. In FIG. 8, the distance between the ship and the infrared digital camera 13 is 452.2 [m], the angle of view is 26 × 20 [degrees], and the sensing element is an SOI diode. In FIG. 8, the tanker's shadow is shown by a solid line, but it is unclearly displayed in a specific photographic image.
[0027]
FIG. 9 shows the classification of ship image data into a horizontal gene array and a vertical gene array in a gene processing step processed by an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention. FIG. 9A shows a gene sequence corresponding to image data, and FIG. 9B shows a state where a horizontal gene sequence and a vertical gene sequence are connected to each other.
[0028]
FIG. 10 shows an infrared image of a container ship representing a divided region based on a gene sequence in which a maximum fitness is calculated in a gene processing step in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention. FIG. 10 is a diagram showing extracted image data as a result after being applied to data and subjected to gene processing, and FIG. 10A shows a photographic image in a state in which a gene sequence is adapted to infrared image data of a container ship, FIG. 10 (a ′) shows an image displayed by a dotted line for a photographic image in which the gene sequence is adapted to the infrared image data of the container ship, and FIG. 10 (b) shows the extracted image data as a result of the gene processing. It is shown. In FIG. 10 (a '), the shadow of the container is indicated by a dotted line, but in an actual photograph, it is unclearly displayed as shown in FIG. 10 (a).
[0029]
FIG. 11 shows an infrared image data of a tanker representing a divided region based on a gene sequence in which a maximum fitness is calculated in a gene processing step in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention. FIG. 11A is a diagram showing the extracted image data obtained after applying the gene processing and the gene processing, and FIG. 11A shows a photographic image in which the gene sequence is adapted to the infrared image data of the tanker. (A ′) shows an image indicated by a dotted line for a photographic image in which the gene sequence is adapted to the infrared image data of the tanker, and FIG. 11 (b) shows extracted image data obtained as a result of the gene processing. It is a thing. In FIG. 11 (a '), the tanker's shadow is shown by a dotted line, but in an actual photograph, it is displayed unclearly as shown in FIG. 11 (a).
[0030]
FIG. 12 is a diagram showing image data as a result of processing infrared image data in the gene processing step in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention. a) is a photographic image of the infrared image data of the container ship, FIG. 12A is a diagram showing the photographic image of the infrared image data of the container ship as an image indicated by a dotted line, and FIG. The image data of the container ship is shown respectively. In FIG. 12 (a '), the shadow of the container ship is indicated by a dotted line, but in an actual photograph, it is unclearly displayed as shown in FIG. 12 (a).
[0031]
FIG. 13 is a diagram showing image data obtained as a result of processing infrared image data in the gene processing step in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention. FIG. 13A shows the tanker infrared image data, FIG. 13A ′ shows the tanker infrared image data in dotted lines, and FIG. 13B shows the processed tanker image data. . In FIG. 13 (a '), the tanker's shadow is shown by a dotted line, but in an actual photograph, it is displayed unclearly as shown in FIG. 13 (a).
[0032]
FIG. 14 is a diagram for explaining the reference line in the reference line calculation processing step in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention.
FIG. 15 is an explanatory diagram for explaining the operation of the reference line calculation processing step in the object type shape discrimination system by the RBF network using the image information according to the embodiment of the present invention. Shows the image data of the actual photographic image, and FIG. 15 (a ′) shows the diagram showing the actual photographic image with dotted lines.
[0033]
FIG. 16 shows image data when the image data obtained in the gene processing step is enhanced by the enhancement processing step in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention. FIG. 16A shows the container ship image data obtained in the gene processing step, and FIG. 16B shows the container ship image data after the enhancement process in the object enhancement process step. It is a thing.
[0034]
FIG. 17 shows image data when the image data obtained in the gene processing step is enhanced by the enhancement processing step in the object type shape discrimination system by the RBF network using the image information according to the embodiment of the present invention. FIG. 17A shows the image data of the car carrier obtained in the gene processing step, and FIG. 17B shows the image data of the car carrier after the emphasis processing in the object emphasis processing step. It is a thing.
[0035]
FIG. 18 shows image data when the image data obtained in the gene processing step is enhanced by the object enhancement processing step in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention. 18 (a) shows the image data of the bulk carrier obtained at the gene processing step, and FIG. 18 (b) shows the image data of the bulk carrier after the enhancement processing at the enhancement processing step. It is a thing.
[0036]
FIG. 19 shows image data when the image data obtained in the gene processing step is emphasized in the structure enhancement processing step in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention. 19 (a) shows the tanker image data obtained in the gene processing step, and FIG. 19 (b) shows the tanker image data after the enhancement processing in the enhancement processing step. It is.
[0037]
FIG. 20 is a diagram in which five types of ships are plotted on the first principal component and the second principal component in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention. The first principal component is taken on the axis, and the second principal component is taken on the vertical axis. Further, in FIG. 20, a circle mark represents a car carrier, a triangle mark represents a tanker, a square mark represents a container ship, a cross mark represents a bulk carrier, and a shape obtained by overlapping two squares represents a fishing boat. .
[0038]
FIG. 21 is a diagram in which two types of ships are plotted on the first principal component and the second principal component in the object type shape discrimination system using the RBF network using image information according to the embodiment of the present invention. The first principal component is taken on the axis, and the second principal component is taken on the vertical axis. Further, in FIG. 21, Δ marks represent tankers and □ marks represent fishing boats.
FIG. 22 is an explanatory diagram showing an RBF (Radial Basis Function) network used in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
[0039]
Next, a specific operation will be described. First, the infrared digital camera 13 in which image data such as a tanker, a container ship, a fishing boat, and an automobile transportation line is collected is connected to the USB board 37 of the computer device 3.
Next, an image take-in command is given from, for example, the mouse 7 to the control processing unit 31 of the computer device 3 executing the object type shape discrimination processing program by the RBF network using the image information. As a result, the control processing unit 31 executes the first step 21 and stores the ship image data in the infrared digital camera 13 in the database area 14 of the hard disk device 33 of the computer device 3.
[0040]
Here, the image data stored in the hard disk device 33 of the computer apparatus 3 from the infrared digital camera 13 is, for example, the image data 300 of the car carrier shown in FIG. 7 or the image data 310 of the tanker shown in FIG.
Next, the control processing unit 31 of the computer apparatus 3 is instructed to move the processing to the next stage using the mouse 7 or the like. Thereby, the CPU 31a of the control processing unit 31 of the computer apparatus 3 executes the second step 22.
[0041]
First, the CPU 31a of the control processing unit 31 of the computer device 3 processes the details of the second step 22 shown in FIG. Here, as described above, the second step 22 includes a gene processing step 221, a reference line calculation processing step 222, an enhancement processing step 223, and a feature parameter processing step 224.
[0042]
When the processing of the second step 22 is started by the control processing unit 31 of the computer device 3, first, the gene processing step 221 is executed by the control processing unit 31. In the gene processing step 221, for example, as shown in FIG. 9A, the image data is converted into the gene sequence “00... 00100... 00100. 010... 00 ”, and the vertical gene sequence“ 010... 010... 010 ”is divided into regions, and this is divided into gene sequences“ 00. ······································································· Connected to “010 ... 010 ... 010” and the divided image data (“00 ... 00100 ... 00100 ... 010 ... “010... 010... 010”), the genetic sequence with the highest fitness is obtained by exploration while repeating the genetic processing, and discriminated and analyzed for each area of the obtained divided image data. The It is to be extracted the 舶 image data.
[0043]
When this series of genetic processing is performed, the image data 330a of the car carrier shown in FIG. 10 (a) and FIG. 10 (a ′) is extracted into the image data 330b of the car carrier shown in FIG. 10 (b). The tanker image data 340a of a) and FIG. 11 (a ′) is extracted into the image data 340b of the tyker shown in FIG. 11 (b), and the image of the automobile carrier shown in FIGS. 12 (a) and 12 (a ′). The data 350a is extracted into the image data 350b of the car carrier shown in FIG. 12B, and the image data 360a in FIGS. 13A and 13A ′ is extracted into the image data 360b shown in FIG. The Specific processing steps of this processing will be described with reference to FIG.
[0044]
That is, when the gene processing step 221 of the image acquisition processing step 22 is executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3, the flowchart of FIG. 5 is executed. First, the CPU 31a of the control processing unit 31 of the computer apparatus 3 takes in, for example, the image data 350a of the car carrier shown in FIGS. 12 (a) and 12 (a ′) from a predetermined area of the database area 14 of the hard disk device 33. Then, the initial gene of the image data 350a is generated (step (S) 221).
[0045]
Specifically, the gene processing step 221 executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3 is performed by converting the image data 350a into, for example, FIG. 9 or FIG. 10 (a) and FIG. As shown in Fig. 9 (a '), the region is divided to obtain the horizontal gene and the vertical gene, and these are connected to generate an initial gene as shown in Fig. 9 (b). (Several 50) is executed (S221). Of course, as shown in FIG. 11 (a) and FIG. 11 (a ′), the initial gene is obtained by dividing the region to obtain the horizontal gene and the vertical gene and connecting them as shown in FIG. 9 (b). May be executed (S2211).
Next, the gene processing step 221 executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3 calculates the degree of applicability by applying Equation 1 to the generated initial gene (S2212).
[0046]
[Expression 1]

[0047]
In Equation 1, the numerator “primary derivative of image data (prewitt operator)” expresses the strength and direction of the contour, and is intended to emphasize the contour of the ship in the image data of the ship. In addition, since the second derivative of the image data (laplacism operator) represents only the strength of the contour, in addition to the contour of the vessel in the image data of the vessel, the noise components in the background and the vessel image are also included. At the same time for emphasis.
[0048]
Next, the gene processing step 221 executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3 executes genetic processing of selection and selection (in this embodiment, surviving the top 50%: 25 individuals). (S2213).
Next, the gene processing step 221 executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3 executes genetic processing such as crossing (in this embodiment, the two-point crossing method: 50 individuals) ( S2214).
Furthermore, the gene processing step 221 being executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3 executes a genetic process such as mutation (in this embodiment, establishment 0.1 [%] (S2215).
[0049]
Further, the gene processing step 221 executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3 obtains the fitness according to the mathematical formula 1 again based on the data after the above-described genetic processing (S2216).
Further, in the gene processing step 221 executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3, the fitness is a constant value over a predetermined generation (for example, 10 generations (see FIG. 6) in this embodiment). If not reached (S2217; NO), the process returns to step 2213 to repeat the above steps 2213, 2214, 2215, and 2216 again.
[0050]
By performing the above processing steps, in this embodiment, as shown in FIG. 6, the solid line indicates the change in fitness of 10 generations when the number of individuals is 30, and the dotted line indicates the number of individuals 50. The change in fitness of 10 generations in the case of individual is shown, and the alternate long and short dash line shows the change in fitness of 10 generations in the case of 70 individuals.
On the other hand, the gene processing step 221 executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3 performs genetic processing such as selection, selection, crossing, and mutation (S2213 to 2215), and the fitness according to Equation 1 is again obtained. When the desired operation is performed (S2216) and the fitness level reaches a certain value over a predetermined generation (for example, 10 generations in this embodiment) (S2217; YES), as shown in FIG. A highly gene sequence is obtained.
[0051]
The gene processing step 221 being executed by the CPU 31a of the control processing unit 31 of the computer device 3 performs binarization based on the gene sequence based on the maximum fitness obtained in the step (S2218). As described above, as a result of the processing performed by the gene processing step 221 being executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3, for example, as shown in FIG. 12B, the car carrier can be clearly displayed. Similarly, for example, as shown in FIG. 13B, the tanker can be clearly displayed.
[0052]
When such a gene processing step 221 is completed, the CPU 31a of the control processing unit 31 of the computer apparatus 3 proceeds to the reference line calculation processing step 222. When the control processing unit 31 starts processing at the reference line calculation processing step 222, for example, the processing shown in FIG. 15 is executed on the image data of the tanker shown in FIG. A process for obtaining the upper deck line is performed on the ship image data.
[0053]
That is, in the reference line calculation processing step 222 processed by the CPU 31a of the control processing unit 31 of the computer apparatus 3, the upper deck line (reference line) SL in the tanker image data 370 shown in FIG. 14 is obtained. The reason for detecting this upper deck line is that it is necessary to separate the hull from the superstructure.
In the reference line calculation processing step 222 processed by the CPU 31a, normally, the position indicating the maximum projection value obtained by projecting the ship image in the horizontal direction is set as the upper deck line (reference line) SL.
[0054]
However, for ship types in which the upper deck line SL exists below the position described above, the reference line calculation processing step 222 is performed as shown in FIGS. 5 (a) and 5 (a ′). An edge is detected by the prewitt operator in the range from the position MP indicating the maximum value of the projection projected in the horizontal direction to the draft position WP (S261 in FIGS. 15A and 15A ′), and thereby the edge image 380b. And thinning processing is performed on the obtained edge image 380b (S262 in FIG. 15A and FIG. 15A ′), thereby obtaining a thinned image 380c. The straight line portion shown in the image 380d obtained by performing the Houh transform on the image 380c is defined as the upper deck line SL (S263 in FIGS. 15A and 15A ′).
[0055]
Thus, based on the upper deck line (reference line) SL obtained in the reference line calculation processing step 222, the emphasis processing step 223 executed by the CPU 31a of the control processing unit 31 of the computer apparatus 3 is performed on the upper deck line. Emphasize the structure. As a result of the processing by the reference line calculation processing step 222, for the container ship image data 380a shown in FIG. 16A, the upper deck line drawn by the container ship (reference line) as shown in FIG. ) The image data 380b is emphasized on the upper side of the SL line.
[0056]
Similarly, for the image data 390a of the car carrier shown in FIG. 17 (a), the upper side of the upper deck line (reference line) SL line drawn on the car carrier is emphasized as shown in FIG. 17 (b). Data 390b is obtained. Similarly, the image data 400a of the bulk carrier shown in FIG. 18A is an image in which the upper side of the upper deck line (reference line) SL line drawn on the bulk carrier is emphasized as shown in FIG. 18B. It becomes data 400b. Similarly, for the tanker image data 410a shown in FIG. 19A, the upper side of the upper deck line (reference line) SL line drawn by the tanker is emphasized as shown in FIG. 19B. become.
[0057]
Next, the CPU 31a of the control processing unit 31 of the computer apparatus 3 proceeds to the process of the feature parameter processing step 224 which is the next step. The characteristic parameter processing step 224 processed by the CPU 31a of the control processing unit 31 is based on the entire ship image data and the image data of the upper structure obtained in the enhancement processing step 223. Characteristic parameters such as the area ratio R, the circularity C, and the height H of the figure centroid.
[0058]
The characteristic parameters calculated in the emphasis processing step 223 are the area ratio R between the entire ship and the superstructure, the circularity C, the height H of the figure center of gravity, and the like.
First, the area ratio R between the entire ship and the superstructure, which is one of the characteristic parameters,
R = Su / Sa ... [2]
Su: Area of superstructure
Sa: Total area of the ship
Given in.
The circularity C, which is another feature parameter, is
C = 4πSc / L²       ... [3]
Sc: Area of ship image after weighting
L: Contour length of ship image after weighting
Given in.
[0059]
Further, the height H of the figure centroid, which is still another feature parameter,
H = (2RY / W) tan (φ / 2) [4]
R: Distance to the other ship
Y: Figure center of gravity in the vertical direction after weighting
W: Total number of pixels in the vertical direction of the image
φ: Camera angle
Given in.
[0060]
In addition, the graphic centroid D in the horizontal direction, which is still another feature parameter,
D = X / Y
X: Weighted figure center of gravity in the horizontal direction
Y: Number of pixels in the horizontal direction of the ship
Given in.
By using the feature parameter obtained in the feature parameter processing step 224 in this manner and processing the third step 23 in the CPU 31a of the control processing unit 31 of the computer apparatus 3, the type of the object (ship) is as follows. And determine the shape.
[0061]
The third step 23 being processed by the CPU 31a of the control processing unit 31 of the computer apparatus 3 is obtained by using the RBF (Radial Basis Function) network 50 shown in FIG. Based on the image data of the superstructure, the area parameters R of the entire ship and the superstructure, the circularity C, and the characteristic parameters of the height H of the figure center of gravity are used as input elements, and the input elements are imported into the RBF network. The RBF network 50 calculates the probability of belonging to each object type, and classifies the types of various objects (ship types) from the probability that the calculation result belongs to a predetermined class.
[0062]
The RBF network 50 can usually determine the types of various objects (ship types) from the attributes described above. For example, as shown in FIG. 20, the container ship and the fishing boat are classified into the same feature range. Further, by calculating the probability based on another input element, the tanker and the fishing boat can be classified as shown in FIG.
[0063]
In order to explain such processing steps, the RBF network 50 uses the predetermined feature parameters (R, C, H) in the first classification 52 to classify the object type (ship type) as shown in FIG. What is classified and classified into the same category in the first classification 52 (tankers, fishing boats), and further classified into the same classification in the second classification 54 using further characteristic parameters (H, D) As an object (classified as a ship type (classified as a tanker or a fishing boat).
However, in the RBF network 50, when the shape of the object can be determined by the probability calculation by the network 50, the first classification 52 and the second classification 54 are not essential requirements.
[0064]
In addition, the area ratio R of the whole ship and the upper structure, the circularity C, and the height H of the graphic center of gravity described above were plotted with five types of ships divided into the first principal component and the second principal component. The thing is a scatter diagram of the main component scores (FIG. 20). As can be seen from FIG. 20, it can be seen that the tankers and fishing boats have a part of the score distribution overlapping.
Therefore, in the discrimination of the ship type in the first stage, it will be explained that both are treated as the same category and the processing proceeds.
[0065]
Next, in order to perform the classification within the same category, the result of the principal component analysis using two characteristic parameters of the height H of the graphic centroid and the horizontal graphic centroid D is shown in FIG. It is. As can be seen from the scatter diagram of FIG. 21, it can be seen that tankers and fishing boats are clearly classified.
In the ship type discriminating system 1 based on this genetic algorithm, the RBF network 50 is used in the third step 23 executed by the computer device 3, and the object type (ship type) is discriminated by the probability calculation by the RBF network 50. Is going.
[0066]
In the object type shape discrimination system 1 by the RBF network using the image information according to the embodiment of the present invention, the RBF network 50 is created based on the result of the principal component analysis described above, as shown in FIG. In the first classification 52, “car carrier ship”, “container ship”, “unloading” are input elements with three characteristic parameters of the area ratio R of the whole ship and the superstructure, the circularity C, and the height H of the center of gravity of the figure. In addition to “ship”, “tanker” and “fishing boat” are classified as the same category, and in the second classification 54, “tanker” and “ It has a two-stage configuration that identifies “fishing boats”. Further, in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention, learning is performed using the feature parameter calculated from the ship image of the ship model for each aspect of 10 degrees.
[0067]
Further, in the object type shape discrimination system 1 by the RBF network using the image information according to the embodiment of the present invention, the vessel discrimination is performed as shown in Equations 2 to 5 described above, such as the distance to the other vessel, the aspect, etc. This is done using infrared images of ships with different conditions.
As described above, in the object type shape discrimination system 1 using the RBF network using image information, the CPU 31a of the control processing unit 31 of the computer apparatus 3 processes the object type shape discrimination processing program 20 using the RBF network using image information. Data as shown in Table 1 was obtained.
[0068]
[Table 1]

[0069]
As shown in Table 1, the discrimination rate is 100 [%] for 11 test images for a car carrier and 75 [%] for 4 test images for a container ship. For bulk carriers, the discrimination rate is 100% for 6 test images, for tankers, the discrimination rate is 100% for 13 test images, and 5 test images for fishing boats. The discrimination rate was 80%.
[0070]
In the object type shape discrimination system 1 using the RBF network using image information, the result processed by the CPU 31a of the control processing unit 31 of the computer device 3 is stored in the database area 16 of the hard disk device 33.
In the object type shape discrimination method by the RBF network using the image information according to the present invention and the system 1 thereof, good results can be obtained for almost all ship types. Moreover, as a result of examining the ship which produced the misclassification, it turned out that the discrimination | determination precision of a ship greatly depends on the processing result of binarization.
[0071]
As described above, the object type shape discrimination method by the RBF network using the image information and the system 1 thereof according to the present invention, the object type (ship type) discrimination using the RBF network 50, almost all object types ( In the ship type), good results can be obtained and athletic performance can be collected. If it is possible to determine the type of such an object, for example, there is an advantage that it is possible to reliably determine the ship type, which is an important element for supporting the safe operation of the ship.
[0072]
As described above, in the present invention, as a preferred embodiment of the object type shape discrimination method by the RBF network using the image information and the system 1 thereof, the ship has been described as an example, but as this object, According to the object type shape discrimination method and system using the RBF network using image information according to the present invention, not limited to a ship, the type and shape of any object such as a car, a tank, a person, etc. are discriminated. Needless to say, that is true.
In the object type shape discrimination method and system using the RBF network using image information according to the present invention, an infrared image is used as an object image. However, the present invention is not limited to this. Even an image obtained by picking up an object by other light using an image pickup means may be used.
[0073]
【The invention's effect】
As described above, according to the object type shape discriminating method and system using the RBF network using the image information according to the present invention, an object image is extracted by a genetic algorithm, and the extracted object image is set to a predetermined reference line. Based on the emphasis of the structure, a feature parameter can be obtained from the emphasized image, and the feature parameter can be input to the RBF network to determine the type and shape of the object.
Further, in the object type shape discrimination method and system using the RBF network using image information according to the present invention, the result of discrimination of the object type can be applied to various technical fields (for example, a collision avoidance system). There is an advantage.
[Brief description of the drawings]
FIG. 1 is a block diagram showing an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
FIG. 2 is a block diagram for explaining a computer device constituting a part of an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
FIG. 3 is a diagram for explaining programs and data stored in a hard disk device in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
FIG. 4 is a diagram showing a configuration example of an object type shape discrimination processing program by an RBF network using image information by an object type shape discrimination system by an RBF network using image information according to an embodiment of the present invention.
FIG. 5 is a specific processing flowchart of gene processing steps processed by a computer device in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
FIG. 6 is a characteristic diagram showing how fitness changes in a gene processing step processed by a computer device in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
FIG. 7 is a diagram showing an example of image data of an automobile carrier ship taken into a computer device from an infrared digital camera in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
FIG. 8 is a diagram showing an example of tanker image data taken into a computer device from an infrared digital camera in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention;
FIG. 9 is a diagram in which ship image data is divided into a gene array in the horizontal direction and a gene array in the vertical direction in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention.
FIG. 10 shows an infrared image of a container ship representing a divided region based on a gene sequence in which a maximum fitness is calculated in a gene processing step in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention. It is a figure which shows the extracted image data of the result after applying to data and carrying out gene processing.
FIG. 11 is a diagram illustrating an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention; It is a figure which shows extraction image data of the result after applying to and carrying out gene processing.
FIG. 12 is a diagram showing image data obtained as a result of processing infrared image data in a gene processing step in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
FIG. 13 is a diagram showing image data as a result of processing infrared image data in a gene processing step in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
FIG. 14 is a diagram for describing a reference line in a reference line calculation processing step in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
FIG. 15 is an explanatory diagram for explaining the operation of the reference line calculation processing step in the object type shape discrimination system by the RBF network using the image information according to the embodiment of the present invention.
FIG. 16 is an image data when the image data obtained in the gene processing step is emphasized in the structure enhancement processing step in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention; FIG.
FIG. 17 shows image data when the image data obtained in the gene processing step is emphasized by the structure enhancement processing step in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention. FIG.
FIG. 18 shows image data when the image data obtained in the gene processing step is emphasized in the structure enhancement processing step in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention. FIG.
FIG. 19 shows image data when the image data obtained in the gene processing step is emphasized in the structure enhancement processing step in the object type shape discrimination system using the RBF network using the image information according to the embodiment of the present invention. FIG.
FIG. 20 is a diagram in which five types of ships are plotted on a first principal component and a second principal component in the object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
FIG. 21 is a diagram in which two types of ships are plotted on a first principal component and a second principal component in the object type shape discrimination system using the RBF network using image information according to the embodiment of the present invention.
FIG. 22 is an explanatory diagram showing an RBF (Radial Basis Function) network used in an object type shape discrimination system using an RBF network using image information according to an embodiment of the present invention.
[Explanation of symbols]
1 Ship classification system using genetic algorithm
3 Computer equipment
5 Keyboard (input device)
7 Mouse (input device)
9 Display (output device)
10 OS storage area
11 Printer (output device)
12 AP storage area
13 Infrared digital camera (imaging means)
14,16 Database area
22 Image acquisition processing steps
24 Ship Image Data Extraction Processing Step
26 Baseline calculation processing step
28 Structure enhancement processing steps
30 Ship type discrimination processing step
31 Control processing unit
31a CPU
31b ROM
31c RAM
32 Main memory
33 Hard disk device
34 I / F for input
35 Display I / F
36 Printer I / F
37 USB board
38 bus line

Claims (7)

An object type determination method capable of determining an object type from image information of an object,
A first step of obtaining image information of the object;
A second step of extracting predetermined feature parameters relating to an image from image information of the object;
Predetermined feature parameters relating to the image obtained in the second step are taken into the RBF network, the probability that each of the object types belongs to the RBF network is calculated, and the type of each object is determined from the probability that the calculation result belongs to the predetermined class. And an object type shape discrimination method using an RBF network using image information. In the second step, the image information of the object is divided into regions based on a genetic sequence to obtain image data, and the highest fitness is achieved while repeating genetic processing on the image data. Gene processing step to obtain the segmented image data by exploring the gene sequence, and to extract the object binarized image data by discriminating and binarizing for each region of the segmented image data,
A reference line calculation processing step for obtaining a reference line for the object image with respect to the object binarized image data extracted in the gene processing step;
An emphasis processing step of emphasizing a predetermined structure based on the reference line in the object binarized image data to obtain emphasized image data;
The image information according to claim 1, further comprising: a feature parameter processing step that calculates a feature parameter related to the image based on the image information of the object and the enhanced image data obtained in the enhancement processing step. Object type shape discrimination method by RBF network. 2. The object type shape discrimination method using an RBF network using image information according to claim 1, wherein the first step obtains image information of the object by imaging the object with an imaging means. In the third step, the feature parameter related to the input image based on the feature parameter related to the image obtained from the image information of the object, the difference between the learning data of the feature parameter related to the image, and the parameter for determining the allowable range The image according to claim 1, wherein an RFB network that calculates a probability that belongs to a predetermined class is used, a feature parameter related to the image is input to the RFB network, and the probability is calculated by the RFB network to determine the type of the object. Object type shape discrimination method by RBF network using information. An object type discrimination system capable of discriminating an object type from image information of an object,
Imaging means for obtaining image information of the object;
Feature extraction means for extracting predetermined feature parameters relating to the image from the image information of the object from the imaging means;
Predetermined feature parameters relating to the image obtained by the feature extraction means are taken into the RBF network, the probability of belonging to each object type is calculated by the RBF network, and the types of various objects are determined from the probability that the calculation result belongs to a predetermined class. An object type shape discriminating system using an RBF network using image information characterized by comprising an object type discriminating means. The feature extraction unit is configured to segment the image information into image data based on a genetic sequence with respect to the image information of the object, and has the highest fitness while repeating the genetic processing on the image data Gene processing means for exploring gene sequences to obtain divided image data, and a gene processing means for extracting object binarized image data by discriminating and binarizing for each region of the divided image data,
Reference line calculation processing means for obtaining a reference line for the object image with respect to the object binarized image data extracted by the gene processing means,
Enhancement processing means for emphasizing a predetermined structure based on the reference line in the object binarized image data to obtain enhanced image data;
6. The image information according to claim 5, further comprising: feature parameter processing means for calculating a feature parameter relating to an image based on the image information of the object and the enhanced image data obtained by the enhancement processing means. Object type shape discrimination system by RBF network. The object type discriminating means includes a feature parameter relating to an input image based on a feature parameter relating to an image obtained from image information of the object, a difference between feature parameter learning data relating to the image, and a parameter for determining an allowable range. 6. The image according to claim 5, comprising an RFB network that calculates a probability belonging to a predetermined class, wherein a feature parameter relating to the image is input to the RFB network, and the probability is calculated by the RFB network to determine the type of the object. Object type shape discrimination system based on RBF network using information.

Images