CN101315701B - Moving Target Image Segmentation Method - Google Patents

Abstract

The invention provides a method for segmenting a moving target image, belonging to the technical field of a computer of image processing; the method comprises the following steps of: adopting a minimum median variance method to obtain a background image; adopting an indirect difference function to obtain a difference image; step 3, selecting a segmentation threshold T and carrying out binarization for the difference image so as to obtaining a binarization foreground image; updating the ground image dynamically according to the updated image, the current background image and the binarization foreground image; using a morphological filtering to eliminate the noise in the binary image and fill defects in the moving target image; the invention can provide help for effective use of a monitoring system and reliable evaluation of monitoring effect, and can be applied in various public places such as police, fire fighting, customs, ports and stations, etc. widely, thus obtaining considerable social effect and improving public safety service.

Description

Moving Target Image Segmentation Method

technical field

The invention belongs to the technical field of image processing computers, and relates to a moving target image segmentation method.

Background technique

Segmentation of moving human body image under complex background is the most basic part of video surveillance image processing, which aims to extract the moving human body area from the background image from the video sequence images obtained by monitoring. Effective segmentation of motion regions is very important for post-processing such as classification, tracking and identification of surveillance objects. However, due to dynamic changes in the background image, such as changes in weather, changes in lighting conditions, background chaos, shadows of moving objects, occlusion between objects and the environment or between objects, and even camera movement, etc. The detection of moving human images becomes a rather difficult task. Therefore, it has aroused the strong interest of many researchers at home and abroad, and has become a frontier hotspot in the field of video image information detection in recent years.

According to the characteristics of the algorithm, the detection methods of moving people in video images under complex backgrounds can be roughly divided into two types of algorithms: motion segmentation and static segmentation. The motion segmentation algorithm uses the motion properties of the moving target to distinguish it from the sequence image background. The static segmentation algorithm first performs static segmentation on a single frame image based on information such as grayscale, texture or contour gradient of each area, and then uses similar block matching method between consecutive frame images for motion estimation, and finally merges each segmented area Complete the extraction of motion targets. For video images with a static background and only simple global scene motion, a differential motion detection algorithm or a background modeling motion detection algorithm is usually used. The differential motion detection algorithm is to threshold the image difference of two adjacent frames or three frames to extract the motion area of the image. Background modeling motion detection algorithm is one of the commonly used motion segmentation methods. Usually, the algorithm first needs to use the complete video image sequence information for statistical modeling to distinguish pixel categories (background/moving area) and generate a background image; then subtract the background from each frame image to obtain the moving target. In short, there are many motion segmentation algorithms, and the key is to find the most appropriate method, that is, to achieve the expected segmentation goal with the least cost.

Contents of the invention

Aiming at the characteristics of static background and panorama motion of video images obtained by fixed-point monitoring in public places, the invention proposes a moving target image segmentation method that can be used in complex background video monitoring. The invention can provide help for the effective use of the monitoring system and the reliable evaluation of the monitoring effect, and can be widely used in various public places such as public security, fire protection, customs, ports, stations, etc., and can obtain considerable social benefits and improve public safety services.

The present invention adopts following technical scheme:

A moving target image segmentation method, each image segmentation adopts the following steps:

Step 1: Obtain the background image using the minimum median variance method;

Step 2: Use the indirect difference function $f(a,b)=1-\frac{2\sqrt{(a+1)(b+1)}}{(a+1)+(b+1)}\frac{2\sqrt{(256-a)(256-b)}}{(256-a)+(256-b)}$ Obtain a difference image, where a and b respectively represent the grayscale (intensity) values of the current image and the background image at the same pixel position, and 0≤a, b≤255, 0≤f(a, b)≤1;

Step 3: Select the segmentation threshold T, $m_{(x,\mathrm{the\; y})}=\left\{\begin{array}{cc}1& f(a_{(x,\mathrm{the\; y})},b_{(x,\mathrm{the\; y})})\&Greater\; Equal;\mathrm{<figure-callout\; id="0"\; label="T"\; filenames="S2008100538305E00012.png"\; state="\{\{state\}\}">T</figure-callout>}\\ 0& \mathrm{otherwise}\end{array}\right.$ Binarize the difference image to obtain a binarized foreground image;

动态更新背景图像；Step 4: Let I _{(x, y)} ^t be the sequence image collected by N frames, B _{(x, y)} ⁿ⁺¹ and B _{(x, y)} ⁿ are the updated and current background images obtained according to step 1 respectively , M _{(x, y)} ⁿ is the binarized foreground image obtained according to step 3, according to the formula

Dynamically update the background image;

Step 5: Use morphological filtering to eliminate noise in the binary image and fill in the absence in the moving target image. Above-mentioned moving object image segmentation method, in step 1, can according to formula $B_{(x,\mathrm{the\; y})}=\arg \left\{{\min }_p{\mathrm{med}}_t{(I_{(x,\mathrm{the\; y})}^t-p)}^2\right\}$ Get the background image of three components of R, G, and B respectively, where p is the color image value to be determined at the pixel position (x, y), and t is the frame index value; in step 3, the method of maximizing the variance between classes can be used To determine the threshold T; in step 4, α can be 0.8.

The video image data used in the invention has the characteristics of static background and panoramic motion, and an algorithm combining dynamic background modeling and differential motion detection is used to segment the moving target image. From the results of each stage in Fig. 3, Fig. 4 and Fig. 5, the overall scheme of the moving body contour image segmentation method in complex background video monitoring adopted by the present invention is suitable; wherein static background modeling, dynamic background update, differential motion Processing links such as segmentation and morphological post-processing are indispensable; the processing flow is basically reasonable, and the final segmentation result is relatively clear and neat.

In the modern social monitoring system, the methods of automatically obtaining the data of the monitoring object can be roughly divided into two categories: one is to use sensors such as piezoelectric, infrared, and ring magnetic induction coils to obtain the parameters of the monitoring object itself, and this method has a higher tracking and recognition rate , but it is easily damaged, and it is also inconvenient to install; there is also a method based on image processing and pattern recognition (the present invention belongs to this kind), which overcomes the limitations of the previous method. Greatly improved, so that the method proposed by the present invention can be implemented. Compared with the former method, the present invention has remarkable technical effect, strong environmental adaptability, long-term stable work, and can monitor without the knowledge of the monitored object, so that the effect of safety monitoring is greatly improved, and can Avoid unnecessary friction and contradictions between the traditional (first type) monitor and the monitored object during monitoring.

Description of drawings

Fig. 1 Segmentation process of moving human body contour image.

Figure 2 Difference functions of different background gray levels.

Figure 3(a) Motion segmentation results of the R component.

Fig. 3(b) Motion segmentation result of G component.

Fig. 3(c) Motion segmentation result of B component.

Figure 3(d) a, b, and c take the "or" operation results.

Figure 4 The results after morphological filtering.

Figure 5. The final moving target segmentation results.

Figure 6 is an example of expansion operation of a binary image, (a) is a binary image, (b) is a structure element B, marked with "+" represents the reference point of the structure element, (c) is the result of expansion.

Figure 7 is an example of corrosion operation of a binary image, (a) is a binary image, (b) is a structural element B, and the mark "+" represents the reference point of the structural element, (c) the corrosion result map.

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

The present invention takes the moving body contour image segmentation method in complex background video monitoring as an embodiment. The whole image segmentation process includes steps such as static background modeling, dynamic background update, differential motion segmentation and morphological post-processing, as shown in FIG. 1 . Each step is further described in detail below in conjunction with embodiments.

1. Static background modeling

The data used in the experiment of the present invention is the gait database released by the Beijing Institute of Automation, Chinese Academy of Sciences. It is a small database of 20 subjects, all collected outdoors with a single camera (Panasonic NV-DX100EN), each subject taking video of walking in three directions (0°, 45°, 90°) sequence data. In the present invention, only the data in the direction of 0° is used, that is, the human side walking sequence.

The present invention uses the least median variance method (Least median of squares, LmedS) to model the background image. The least median variance method (LmedS) is an algorithm proposed based on the theory of robust statistics. The concept of robustness refers to the degree of influence of individual outliers on statistics. Robust statistics is a parameter estimation method suitable for outliers in the sample group. Robust statistics have received a lot of attention in the field of computer vision, mainly because the input data of computer vision problems are often disturbed by outliers.

In robust statistics, in order to evaluate the ability of an algorithm to resist outlier disturbances, Hampel proposed the concept of BP (Breakdown Point). Since it is an asymptotic result at first, it is inconvenient to calculate, so Donoho and Huber defined the BP for finite samples as:

Where T is the estimated value of the parameter, n is the content of the sample X, and β(m; T, X) represents the supremum of the difference between the estimated values of the parameters on both sides after any replacement of m points in X. For BP with limited samples, it means the minimum proportion of the number of outliers in the data that will not invalidate the estimated value when using this method for parameter estimation. When the proportion of outliers in the data exceeds BP, the estimated value will become very unstable.

There are basically two ways to deal with abnormal points, one is the self-coordination method (Accommodation-based approach), that is, the method itself can withstand the interference of abnormal points, and the other is to eliminate the abnormal points first, and then deal with them according to the traditional method. The least median variance algorithm (LmedS) is a combination of the two. Under the premise of self-coordination, the least mean square estimation algorithm (Least mean square, LMS) is used for estimation. Therefore, it can eliminate the interference of abnormal points and obtain a comparative Good estimated effect. Rousseeuw and Leroy's definition of the minimum median variance method is:

则：In the known sequence X={x ₁ , x ₂ ,... _xi ,...x _N }, x _i is the N observed values of x, and x is estimated according to x _i , assuming that the estimated value is

but:

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; min</span>}}_{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}{\mathrm{<span\; class="notranslate">\; med</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

where i=1, 2...,N.

The video sequences in the database used in the present invention are captured by fixed cameras. In theory, if there is no interference or other influences at all, the background is still. For single-frame analysis of the video sequence, track the curve of the gray level at a certain point over time, and the curve should basically remain stable. However, for points where there are moving objects or other environmental interference factors, the gray distribution curve will change with time. The present invention attempts to track and observe the gray levels of pixels at different positions in the gait sequence.

The stationary background modeling process is described below.

If let I _{(x, y)} ^t represent the sequence images collected by N frames, where t represents the frame index value (t=1, 2,., N), (x, y)∈I ^t , then the background B _{(x, y)} is:

${\mathrm{<span\; class="notranslate">\; B</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; min</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}{\mathrm{<span\; class="notranslate">\; med</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

In the formula, p is the value of the color image (R, G, B) to be determined at the pixel position (x, y). If each component is an 8-bit image, the value range of p is 0-255; t is the frame Index value, which varies between 1 and N. (arg represents the value of the unknown number that satisfies the requirements in ()) The specific process of the algorithm is (only taking the R component as an example):

(i) select a pixel point position (x, y);

(ii)p=0;

(iii) sequentially calculate (I _{(x, y)} ¹ -p) ² , (I _{(x, y)} ² -p) ² , ..., (I _{(x, y)} ^N -p) ² ;

(iv) Sort the calculation results. If N is an even number, take the average of the N/2th and (N+1)/2 numbers after sorting. If N is an odd number, take the N/2th number and save the result To the array med, namely med ₀ ;

(v)p=p+1, when p<=255, return to (iii), repeat (iii), (iv), (v), save the result as med _p , otherwise execute (vi);

(vi) Find the minimum value among med ₀ , med ₁ , ... med ₂₅₅ , and the corresponding p value is the background gray value of the pixel position.

(vii) Reselect the pixel point position, return to (ii) and repeat until all the pixel points in the image are calculated.

Considering that the database images are all in RGB format, the three components of R, G, and B are modeled separately here, and color background images in R, G, and B formats can also be obtained after synthesis.

1. Differential Motion Segmentation

In order to determine the moving target, the most commonly used method is to perform threshold segmentation on the difference image obtained by subtracting the current image from the background model. A big disadvantage of this method is that for low-contrast images, it will be difficult to determine the segmentation threshold because of the small grayscale change, that is, it is difficult to clearly extract the moving target from the background. For this reason, the present invention uses an indirect difference function instead to perform the difference operation. The expression of the difference function is:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\sqrt{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\sqrt{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

Where a and b respectively represent the grayscale (intensity) values of the current image and the background image at the same pixel position, and 0≤a, b≤255, 0≤f(a, b)≤1. When a=b, f(a, b)=0; when a and b are different, f(a, b) is proportional to the difference between a and b. At the same time, the sensitivity of the difference function can be changed automatically with the background gray value. Taking b=5 and b=100 as an example, the difference function is shown in FIG. 2 . It can be seen from the figure that b is different, and the difference function f(a, b) is also different. When b is small (b=5), the difference function becomes larger rapidly with the growth of a, which shows that the difference The sensitivity of the function is automatically increased, and this adaptability improves the accuracy of image segmentation.

After determining the difference value between the background image and the current image, it is necessary to select a segmentation threshold T (0≤T≤1). The present invention adopts the Otsu method (the paper source: Otsu N.A threshold selection method from gray-level histograms. In: IEEE Trans. Systems, Man and Cybernetics, 1979, SMC-9 (1), 62~66), that is, by maximizing the Variance to determine the threshold T. Its binarization process can be expressed as:

${\mathrm{<span\; class="notranslate">\; m</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; <figure-callout\; id="0"\; label="T"\; filenames="S2008100538305E00012.png"\; state="\{\{state\}\}">T</figure-callout></span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

2. Dynamic background update

Although the static background model is established by the minimum median variance method above, due to the influence of noise and illumination changes, the background of the actual video sequence does not always remain static. In order to obtain a more accurate differential motion segmentation (ie, background subtraction) effect, the background must be updated dynamically. The present invention adopts the Kalman filtering method of Karmann and Brandt to carry out the dynamic updating of background (the paper source: Karmann K, Brandt A. Moving object recognition using an adaptive background memory.In: Cappellini V ed. Time-varying Image Processing and Moving Object Recognition.2 . Elsevier, Amsterdam, The Netherlands, 1990). Let B _{(x, y)} ⁿ⁺¹ and B _{(x, y)} ⁿ be the updated and current background respectively, according to the binarized foreground image obtained by the motion detection in the current frame I _{(x, y)} ⁿ M _{(x, y)} ⁿ , the background update process is:

Wherein α is a weighting coefficient, and the present invention finds that 0.8 is more suitable for α through experimental analysis. When the updated background is close enough to the pre-updated background, stop updating.

The results of motion segmentation with automatically updated background are shown in Fig. 3.

3. Morphological post-processing

There will inevitably be noise in the image after motion segmentation, and at the same time, a small number of points in the moving target will be misjudged as the background, so the image needs to be post-processed to obtain the best segmentation effect. The present invention uses morphological filtering to eliminate noise in binary images and fill in the absence of moving objects.

In morphology, dilation operation and erosion operation are the most basic morphological transformations.

① Dilation

表示，X用B来膨胀记为

定义为The expansion operation is also called the expansion operation, with the symbol

Indicates that X is expanded by B and denoted as

defined as

The expansion process can be described as follows: set B first makes a mapping B^ about the origin, then translates x to form a set (B^)x, and finally calculates the set of structural element reference points where the set (B^)x and the set X are not empty sets . In other words, the set obtained by expanding X with B is the set of reference point positions of structural element B when the displacement of B^ intersects with at least one non-zero element of set X.

Example 1. Example of dilation operation

与X相比发现，X按照B的形态膨胀了一定范围。因此，该运算被名之为膨胀。As shown in Figure 6(a) is a binary image, the shaded part represents the area with high gray value (generally 1), the white part represents the area with low gray value (generally 0), the spatial coordinates of the upper left corner is (0, 0). (b) The picture shows structural element B, marked with "+" to represent the reference point of the structural element. The result of the expansion is shown in figure (c), where the black part is the expanded part. put the result

Compared with X, it is found that X expands to a certain extent according to the form of B. Hence, the operation is called dilation.

② Corrosion operation (Erosion)

定义为The corrosion operation is also called the erosion operation, with the symbol Indicates that X is corroded by B and denoted as

defined as

The erosion process can be described as follows: the collection of structural element reference points that are still in the set X after the translation of the set B by x. In other words, the set obtained by corroding X with B is the set of reference point positions of B when B is completely included in set X.

Example 2 Corrosion operation example

As shown in Figure 7 (a) is a binary image, (b) is the structural element B, marked with "+" represents the reference point. The result of corrosion is shown in Figure (c), where the black part is the part left after corrosion. put the result Compared with X, it is found that the area of X has been reduced, and it can be seen that the parts that cannot accommodate structural elements have been corroded.

In morphology, the opening operation AοB refers to the result of A being corroded by B and then expanded by B, namely:

表示腐蚀运算，

表示膨胀运算。开运算可完全删除不能包含结构元素的对象，如平滑对象的凸轮廓、断开狭窄的连接、去掉细小的突起部分。闭运算与开运算刚好相反，它的定义是指A被B膨胀后再用B来腐蚀的结果，即：In the formula

represents the erosion operation,

Indicates the dilation operation. The opening operation can completely delete objects that cannot contain structural elements, such as smoothing the convex contour of objects, breaking narrow connections, and removing small protrusions. The closing operation is just the opposite of the opening operation. Its definition refers to the result of A being expanded by B and then corroded by B, namely:

Closing operations can fill holes smaller than structural elements, such as smoothing the concave contours of objects, and connecting slits into slender bends. The functions of filtering and filling holes can be realized by using the above properties of the opening and closing operations. Figure 4 shows the results of morphological filtering.

After morphological filtering, the noise is not necessarily completely eliminated. Some stray noise may form blocks of different sizes, and the moving target is often the largest of these blocks. Therefore, the connected domain analysis of the image can be carried out, aiming at Only moving objects are kept in the image. The steps of connected domain analysis are:

(i) mark the connectivity matrix;

(ii) Calculate the number of pixels in each matrix;

(iii) find the matrix with the most pixels;

(iv) Determine the goal of the exercise.

The final segmentation of moving objects is shown in Fig. 5.

Claims (4)

1. movement destination image partition method, each image segmentation adopts following step:

Step 1: adopt minimum meta variance method to obtain background image;

Step 2: adopt indirect difference function $f(a,b)=1-\frac{2\sqrt{(a+1)(b+1)}}{(a+1)+(b+1)}\frac{2\sqrt{(256-a)(256-b)}}{(256-a)+(256-b)}$ Obtain difference image, a wherein, b represent present image and background image gray scale (intensity) value in same pixel position respectively, and 0≤a, and b≤255,0≤f (a, b)≤1;

Step 3: select segmentation threshold T, $M_{(x,y)}=\left\{\begin{array}{cc}1& f(a_{(x,y)},b_{(x,y)})\&GreaterEqual;T\\ 0& \mathrm{Otherwise}\end{array}\right.$ With the difference image binaryzation, obtain the binaryzation foreground image;

Step 4: establish I _{(x, y)} ^tBe the sequence image that the N frame is gathered, B _{(x, y)} ^N+1And B _{(x, y)} ⁿBe respectively after the renewal that obtains according to step 1 and current background image, M _{(x, y)} ⁿFor the binaryzation foreground image that obtains according to step 3, according to formula Wherein, α is a weighting coefficient, dynamically updates background image;

Step 5: utilize morphologic filtering to eliminate the noise in the bianry image and fill up disappearance in the movement destination image.

2. movement destination image partition method according to claim 1 is characterized in that, in the step 1, according to formula $B_{(x,y)}=\arg \left\{{\min }_p{\mathrm{med}}_t{(I_{(x,y)}^t-p)}^2\right\}$ Obtain the background image of R, G, three components of B respectively, p is that (x y) locates coloured image value to be determined to location of pixels, and t is the frame index value in the formula.

3. movement destination image partition method according to claim 1 is characterized in that, adopts maximization inter-class variance method to determine threshold value T in the step 3.

4. movement destination image partition method according to claim 1 is characterized in that in the step 4, α gets 0.8.

Images