JP5288440B2 - Human body detection apparatus and human body detection method - Google Patents

Description

  The present invention relates to a human body detection device and a human body detection method used for detecting, for example, that a person has entered a facility in facility security.

As the above-described human body detection device, for example, a far-infrared camera as an imaging unit that captures a thermal image and a distance to an imaging target as a human candidate extracted from the temperature distribution of the thermal image obtained by the far-infrared camera. Based on the distance measuring means to be measured and the distance to the imaging target measured by this distance measuring means, a candidate is selected and a template suitable for the imaging target is selected from a large number of templates created for each distance. Some of them have a signal processing unit that performs a person extraction process such as template matching (see, for example, Patent Documents 1 and 2).
JP 2006-101384 A JP 2006-153777 A

  However, in the above-described conventional human body detection device, paying attention to the fact that a person has a temperature higher than that of the background in principle, a candidate for a high-temperature part in the temperature distribution of a thermal image acquired by a far-infrared camera Therefore, there is a possibility that a person cannot be detected if the face or arms are not exposed or if the background temperature is high during the daytime.

In addition, since it is necessary to create a template for the brightness value and the contour for each distance using the distance to the imaging target as a parameter, that is, it is necessary to create a large number of templates. It has been a conventional problem to solve these problems.
The present invention has been made by paying attention to the above-described conventional problems, and can detect a person not only in a low-temperature environment such as nighttime but also in a high-temperature environment such as summer daytime and the cost of creating a template. It is an object of the present invention to provide a human body detection device and a human body detection method capable of significantly reducing the above.

  Here, the person and the background naturally have a temperature difference due to the difference in each material and heat capacity. If there is a temperature difference between the two, this temperature difference can be extracted as an edge on the thermal image, and the contour shape (edge) of a person is particularly characteristic. In the present invention, in detecting the human body, not focusing on the high temperature of the body temperature as in the past, but focusing on the peculiarity of the contour shape of the person, creating a line drawing based on the edge extracted on the thermal image, A human body is detected by performing template matching on the line drawing.

In addition, since the contour shape has similarity according to the distance, the human template used in template matching has one motion pattern and one posture according to the farthest position within a predetermined range where the human body needs to be detected. One sheet was created for each pattern.
In other words, a plurality of types of humanoid templates having the same ratio are created. At this time, the humanoid template includes a line thicker than the line drawing created based on the edge (for example, the line drawing line created based on the edge). Using a line about three times thicker than the above, or a line that overlays multiple patterns of images close to silhouettes when moving parts such as arms and legs move (moving parts appear thicker) The fine irregularities of the line drawing can be absorbed.

Then, when performing template matching using such a humanoid template, the acquired thermal image is reduced based on the distance to the imaging target in this image so as to match the size of the humanoid template. .
That is, the invention according to claim 1 of the present invention includes an imaging unit (for example, a far infrared camera) that captures a thermal image, a distance measuring unit that measures a distance to an imaging target in the thermal image captured by the imaging unit, A line drawing creation filter that creates the line drawing of the imaging target from an edge extracted by a temperature difference in the thermal image, and a line drawing creation filter that is created in accordance with the farthest distance within a predetermined range, is created by the line drawing creation filter, and A signal processing unit including a plurality of types of human-type templates that perform matching on the line drawing of the imaging target reduced based on the distance to the imaging target measured by the distance measuring unit; The template of the imaging target is checked at a template matching stage in which a template template is compared with the line drawing of the imaging target created by the line drawing creation filter. Create a bold line than該線image lines in order to absorb picture of the fine irregularities, the signal processing unit has a noise filter, the noise filter is provided with a search frame that is formed by arranging a predetermined number of pixels In the step of searching the luminance value of the outer edge portion by overlaying the search frame on an unknown line whether or not necessary for human body detection, when the unknown line does not cross a pixel located at the outer edge portion, The unknown line is removed as a line unnecessary for human body detection, and the signal processing unit calculates an amount of the line drawing of the imaging target that overlaps the thick line of the human template using an evaluation function, and the calculated value is a threshold value. It is characterized in that the imaging object is detected as a human body when exceeding the above, and the configuration of the human body detection device is a means for solving the above-described conventional problems.

Further, in the human body detection device of the present invention, as claimed in claim 2, the lines of the humanoid template in the signal processing unit are approximate to each other when the line drawing of the imaging target created by the line drawing creation filter forms a human figure overlaps the humanoid plurality patterns, i.e., that have a structure that a thickness of conjugated multiple patterns that approximate.

On the other hand , in the human body detection method according to claim 3 of the present invention, when the human body is detected by the human body detection device according to claim 1 or 2 , the distance measurement unit is configured to capture the thermal image by the imaging unit. After measuring the distance to the imaging target in the thermal image, the signal processing unit creates a line drawing of the imaging target with a line drawing creation filter and sets the distance to the imaging target to match the size of the human template. Based on this, the line drawing is reduced, and then a line unnecessary for human body detection is removed from the reduced line drawing to be picked up by a noise filter having a search frame. The amount of the line drawing to be imaged that overlaps the thick line of the human template is calculated using an evaluation function, and the calculated value exceeds the threshold value. Wherein the imaging target has been characterized by being configured to detect a human body, and the structure of the human body detection method as a means for solving the conventional problems described above when the.

  The evaluation function used in the template matching of the present invention is to calculate the number of pixels on the thick line of the humanoid template by comparing the image of the humanoid template with the created line drawing (step of FIG. 5). For the characteristics of such an evaluation function, if the above-described noise filter is used to remove lines unnecessary for human body detection, the value of the evaluation function in the background is lowered, The separation of the person and the background on the evaluation function can be promoted.

In the human body detection apparatus according to the present invention, as a distance measurement unit that measures the distance to the imaging target in the thermal image captured by the imaging unit, for example, a laser range finder or a stereo camera can be used, and millimeter waves or echoes can be used. Measurement means can be used.
Further, in the human body detection device according to the present invention, for example, when the vehicle is mounted on a vehicle using a laser range finder as a distance measuring means, the vehicle moves around a corner or travels with an uneven surface. As a roll angle (tilt) is generated, a part of the scan line of the laser range finder may deviate from the assumed human detection range, so attitude measurement such as a gyroscope that measures the attitude of the vehicle It is desirable to equip the means.

  For example, if a gyroscope is equipped, and the roll angle (tilt) occurs in the vehicle, the height of the measurement point is calculated in the signal processing unit from the distance measured by the laser range finder and the attitude of the vehicle by the gyroscope. Based on the calculated height, it is determined that the distance measurement value of the area is not used.

In the human body detection device according to claim 1 or 2 of the present invention and the human body detection method according to claim 3 , since it is configured as described above, not only in a low temperature environment such as nighttime but also in a high temperature environment such as summer daytime. In addition to being able to detect people and cover a line drawing of multiple images per humanoid template, it is extremely superior that the cost of creating a template can be greatly reduced. The effect is brought about.

Further, the human body detecting method according to the human body detecting device and claim 3 according to claim 1 or 2 of the present invention, in addition to effects as described above can be obtained, also be implemented to improve the human detection performance This is a very good effect.

Hereinafter, a human body detection device and a human body detection method according to the present invention will be described with reference to the drawings.
1 to 5 show an embodiment of a human body detection device according to the present invention. In this embodiment, a case where the human body detection device is mounted on a mobile robot will be described as an example.
As shown in FIGS. 1 and 2, the human body detection device 1 mounted on the mobile robot R includes a far-infrared camera 2 as an imaging unit that captures a thermal image, and imaging in the thermal image captured by the far-infrared camera 2. Thermal image data from a laser range finder 3 as a distance measuring means for measuring the distance to an object existing in the area including the targets T1 and T2, a gyroscope 4 for measuring the attitude of the mobile robot R, and the far-infrared camera 2 , A signal processing unit 10 for processing the distance information from the laser range finder 3 and the attitude signal from the gyroscope 4 is provided.

The signal processing unit 10 creates a line drawing creation filter 11 that creates a line drawing including the imaging targets T1 and T2 from the edges extracted by the temperature difference in the thermal image taken by the far infrared camera 2, and the line drawing creation filter 11 creates the line drawing. Then, a plurality of types of humanoid templates 12 for matching the line images s1 and s2 including the imaging targets T1 and T2 to be reduced as will be described later are provided.
In this case, the human template 12 is created one by one corresponding to the motion pattern and posture pattern in accordance with the farthest distance within a predetermined range, and this human template 12 is obtained by the laser range finder 3. A line image cut out based on the correspondence between the scanning phase of the laser range finder 3 and the angle of view of the far-infrared camera 2 from the line images corresponding to the regions A1 and A2 selected as the human presence region candidates on the measured distance difference profile. Matching to line drawings s1 and s2 including the imaging targets T1 and T2 obtained by reducing S1 and S2 based on the distances L1 and L2 to the imaging targets T1 and T2 measured by the laser range finder 3 Is supposed to do.

  In this embodiment, the human-type template 12 is created by the line drawing creation filter 11 and is compared with the line drawings s1 and s2 including the reduced imaging targets T1 and T2 at the stage of template matching. The lines are thicker than the lines s1 and s2 in order to absorb the unevenness. More specifically, silhouettes such as movement patterns such as the positions of arms and legs and posture patterns that hold or carry a heel. It is created with thick lines that can absorb (overlapping with) the line drawing that forms a human figure with a plurality of similar patterns.

Further, as shown in FIG. 3A, the signal processing unit 10 determines whether or not the line drawings s3 and s2 including the imaging targets T3 and T2 created and reduced by the line drawing creation filter 11 are necessary for human body detection. When there is an unknown line UL, a noise filter 13 is provided that can remove the unknown line UL from the line drawings s3 and s2, as shown in FIG.
As shown in FIG. 4A, the noise filter 13 includes a search frame 14 formed by arranging a predetermined number of pixels P, and the search frame 14 is placed on a line UL where it is unknown whether human body detection is necessary. In the stage of searching for the luminance value of the outer edge by superimposing, as shown in FIG. 4B, when the luminance values of the pixels PE located at the outer edge are all zero (the unknown line UL is the outer edge). 4), the unknown line UL is removed as a line unnecessary for human body detection. On the other hand, as shown in FIG. 4C, the luminance value of the pixel PE located at the outer edge is removed. Is recognized in either case (when the unknown line UL crosses the pixel PE located at the outer edge), the unknown line UL is left as a line necessary for human body detection.

Next, the human body detection procedure by the above human body detection apparatus 1 will be described.
First, as shown in FIG. 5, after the thermal image is taken by the far-infrared camera 2 in step St1, the luminance is normalized in step St2, and then the sensitivity of the far-infrared camera 2 is appropriate in step St3. If so, the process proceeds to step St5 without changing the gain. If it is necessary to adjust the sensitivity of the far-infrared camera 2, the gain is adjusted in step St4.

Next, in step St5, a process using an intermediate value filter is performed to remove noise in the thermal image, and in step St6, the line drawing creation filter 11 creates line drawings S1 and S2 including the imaging targets T1 and T2.
On the other hand, the laser by range finder 3 to measuring distance meter range including the imaging object T1, T 2 in the thermal image in step St 7, after the creation of the measured distance profile with respect to the scanning phase of the laser range finder 3 in step St8, In step St9, an edge is extracted from the measurement distance profile. When extracting the edge, the posture measurement result from the gyroscope 4 obtained in step St10 is reflected.

In step St11, the scanning phase is divided into areas A1 and A2 with the edge as a boundary. In step St12, average values of the distances of the areas A1 and A2 are calculated, respectively, and are used as representative distances of the areas A1 and A2. Reflected in St13.
Next, in step St13, the thermal image is divided based on the correspondence between the scanning phase and the angle of view of the laser range finder 3. In step St14, the representative distances of the regions A1 and A2 measured by the laser range finder 3 and the human template The line drawings S1 and S2 including the imaging targets T1 and T2 located in the areas A1 and A2 are reduced based on the ratio to the creation distance of 12, and line drawings s1 and s2 are obtained.

Subsequently, in step St15, the noise filter 13 is processed on the line drawings s1 and s2 including the imaging targets T1 and T2 reduced in step St14 to remove unknown lines UL that are not necessary for human body detection.
Then, as shown in FIG. 6, the human template 12 is selected and collated with the human template 12 for the line drawing s1 of the line drawings s1 and s2 in steps St16 to St18, and calculated by the evaluation function in step St19. The value (the amount of the line drawing s1 overlapping the thick line of the humanoid template 12) is compared with the threshold value. When the value calculated by the evaluation function exceeds the threshold value, the imaging target T1 is detected as a human body in step St20. The position of the imaging target T1 is calculated.

Next, returning to step St16, the processing of steps St16 to 18 is performed for the line drawing s2 of the line drawing s1 and s2, and when the value calculated by the evaluation function exceeds the threshold value in step St19, the imaging target T2 is set in step St20. Is detected as a human body, and the position of the imaging target T2 is calculated.
During this time, when the calculated value by the evaluation function does not exceed the threshold value in step St19, the process returns to step St17 to select another human template 12, and thereafter the processing of steps St17 to 19 until the calculated value by the evaluation function exceeds the threshold value. Repeat.

  In the human body detection device 1 described above, the humanoid template 12 of the signal processing unit 10 is created one by one corresponding to the motion pattern and posture pattern in accordance with the farthest distance within a predetermined range. By making the line of the template 12 thicker than the lines of the line drawings s1 and s2 including the imaging targets T1 and T2, it is more concrete so that the fine irregularities of the line drawings s1 and s2 can be absorbed at the template matching stage. Because it is designed to absorb line patterns that form multiple human figures with similar silhouettes, such as movement patterns such as the positions of arms and legs and posture patterns that hold or carry a heel, a single human figure Since the template 12 can cover the line drawings s1 and s2 including the plurality of imaging targets T1 and T2, the production cost of the template 12 can be greatly reduced. And thus to be.

  Further, in the human body detection device 1 described above, when the noise filter 13 is provided in the signal processing unit 10, there is an unknown line UL that is not necessary for human body detection in the line drawings s 1 and s 2 including the imaging targets T 1 and T 2. Since the unknown line UL is removed from the line drawings s1 and s2, the value of the evaluation function in the background is lowered, and separation of the person and the background on the evaluation function is promoted, and as a result Thus, the human detection performance is improved.

  In the above-described embodiment, the case where the laser range finder 3 is used as a distance measuring unit that measures the distance to an object within the range including the imaging targets T1 and T2 in the thermal image captured by the far-infrared camera 2 is shown. However, the present invention is not limited to this, and as another configuration, for example, as shown in FIG. 7, a stereo camera 23 arranged in accordance with the width of the mobile robot R can be used as the distance measuring means.

  In the above-described embodiment, the case where the human body detection device 1 is mounted on the mobile robot R is shown, but the present invention is not limited to this.

  The human body detection device and the human body detection method according to the present invention can be used, for example, to detect the intrusion of a person into a facility in the facility security, and to be mounted on a robot that supports and supports a guard. Can be used to recognize guards acting together.

1 is an explanatory perspective view schematically showing a mobile robot equipped with a human body detection device according to an embodiment of the present invention. FIG. 2 is a schematic explanatory diagram of signal processing such as image division and image reduction by a signal processing unit of the human body detection device in FIG. 1. FIG. 2A is a line drawing explanatory diagram before a noise filter process showing a function of a noise filter included in a signal processing unit of the human body detection apparatus in FIG. 1 and FIG. It is explanatory drawing (a)-(c) which shows the criteria of a line unnecessary for the human detection of the search frame which the noise filter in FIG. 3 has. It is the first half part of the flowchart at the time of performing human body detection by the human body detection apparatus in FIG. It is the latter half part of the flowchart at the time of performing a human body detection by the human body detection apparatus in FIG. It is a perspective explanatory view which shows simply the mobile robot carrying the human body detection apparatus by other embodiment of this invention.

Explanation of symbols

1 Human body detection device 2 Far-infrared camera (imaging means)
3 Laser range finder (distance measuring means)
DESCRIPTION OF SYMBOLS 10 Signal processing part 11 Line drawing creation filter 12 Humanoid template 13 Noise filter 14 Search frame 23 Stereo camera PE Pixel located in outer edge part S1, S2 Line drawing s1, s2 Reduced line drawing T1, T2 Imaging object UL Unknown line
T (ξ, η) Template image brightness value
I (x + ξ, y + η) Template matching target image brightness value
x, y template matching target image coordinates
ξ, η template image coordinates

Claims (3)

Imaging means for capturing a thermal image;
Distance measuring means for measuring the distance to the imaging target in the thermal image captured by the imaging means;
A line drawing creation filter that creates the line drawing of the imaging target from an edge extracted by a temperature difference in the thermal image, and a line drawing creation filter that is created in accordance with the farthest distance within a predetermined range, is created by the line drawing creation filter, and A signal processing unit including a plurality of types of humanoid templates that perform matching on the line drawing of the imaging target reduced based on the distance to the imaging target measured by the distance measuring unit;
Lines of the line drawing are absorbed at the stage of template matching in which the human template of the signal processing unit is compared with the line drawing of the imaging target created by the line drawing creation filter. Create a thicker line,
The signal processing unit includes a noise filter, and the noise filter includes a search frame formed by arranging a predetermined number of pixels, and the search frame is superimposed on a line that is unclear whether it is necessary for human body detection. In the step of searching for the luminance value of the outer edge, when the unknown line does not cross a pixel located at the outer edge, the unknown line is removed as an unnecessary line for human body detection,
The signal processing unit calculates an amount of the line drawing of the imaging target that overlaps a thick line of the human template with an evaluation function, and detects the imaging target as a human body when the calculated value exceeds a threshold value. A human body detection device.   The line of the humanoid template in the signal processing unit has a thickness that overlaps the line drawing forming the humanoid having a plurality of patterns that are similar to each other when the line drawing to be imaged created by the line drawing creation filter forms a human figure. The human body detection device according to claim 1. When performing human body detection by the human body detection device according to claim 1 or 2 ,
After taking the thermal image by the imaging means, after measuring the distance to the imaging target in the thermal image by the distance measuring means,
In the signal processing unit, the line drawing creation filter creates the line drawing of the imaging target, and reduces the line drawing based on the distance to the imaging target to match the size of the human template,
Next, after removing a line unnecessary for human body detection from the reduced line image to be imaged by a noise filter having a search frame, template matching is performed by comparing the line image with the human template, A human body detection method using a human body detection device, wherein an amount of a line drawing of the imaging target that overlaps a thick line is calculated by an evaluation function, and the imaging target is detected as a human body when the calculated value exceeds a threshold value.

Images