CN106934819A - A kind of method of moving object segmentation precision in raising image - Google Patents

Abstract

The invention belongs to technical field of image processing, a kind of method for improving moving object segmentation precision in image is especially provided, each two field picture to gathering carries out pixel level picture steadiness and judges, if stabilization, image procossing is then carried out using Vibe algorithms, otherwise, image procossing is carried out using frame differential method, morphology filling treatment finally is carried out to the image after the image after frame differential method is processed or Vibe algorithm process；Improve the defect of Vibe algorithms presence using frame differential method and morphology filling, the real-time detection precision of moving object in image can be improved.

Description

A Method of Improving the Detection Accuracy of Moving Objects in Images

technical field

The invention belongs to the technical field of image processing, in particular to a method for improving the detection accuracy of moving objects in images.

Background technique

At present, intelligent monitoring technology is playing an increasingly important role in the security monitoring of companies and enterprises, in the traffic management of society, and in people's life and entertainment.

Intelligent monitoring technology has important research significance in both theory and practical application. For example, from a security point of view, intelligent surveillance video can effectively analyze video content and extract potentially dangerous targets in the video. Examples include possible robbers in a bank, or other moving targets in a surveillance location. Reporting these warnings to security supervisors can effectively improve the efficiency of video surveillance and save costs related to security supervision. From the perspective of traffic management, intelligent video technology can intelligently analyze the monitored road conditions, identify the degree of congestion on the monitored roads, or monitor and track illegal driving vehicles, all of which can help traffic managers to conduct traffic management more effectively. In terms of people's leisure and entertainment, it is possible to recognize human body movements through intelligent video, interact with the game system according to the recognized movements, and recognize human body movements through cameras to complete the control of the game.

The moving target detection in intelligent monitoring refers to detecting the changing area in the sequence image and extracting the moving target from the background image. key technologies. The motion information in the image can be obtained through the detection of the moving object, and the moving object in the image can be extracted and positioned, thus simplifying the difficulty of subsequent motion tracking, identification, and analysis.

Traditional moving target detection methods mainly include three types: background difference method, frame difference method, and optical flow method. In recent years, the Vibe algorithm in the background difference method has been paid more and more attention.

The Vibe algorithm is a pixel-level foreground detection (or background modeling) method. Its basic idea includes: storing a sample set for each pixel, which contains the past pixel values of this pixel and its surrounding neighbors. When a new pixel is encountered, the pixel value of the new pixel is compared with the collection points in the sample set to determine whether the new pixel is a background point. This algorithm is a process of classifying background points. The Vibe algorithm has the advantages of high computing efficiency, good implementation effect, less memory usage and optimal sample attenuation. However, the Vibe algorithm has ghosts caused by sudden changes in factors such as lighting and shadows in the environment (wrongly detecting the area covered by moving objects in the previous few frames as moving objects), and missed detection (meaning that some moving objects are not detected in image acquisition. )And other issues. Moreover, when the Vibe algorithm detects moving objects, the detection of moving objects will appear empty (meaning that the object itself is moving during detection, but some parts of the moving object are still when detected). As shown in Figure 3, the whole person is originally in motion, so the whole person should be white, but the part of the person's head that is black during detection is a hollow phenomenon.

Contents of the invention

In order to solve the shortcomings of the traditional Vibe algorithm, the present invention proposes a method for improving the motion detection accuracy in images.

The present invention is a method for improving the detection accuracy of a moving object in an image. The pixel-level image stability is judged for each frame of image collected. If it is stable, the Vibe algorithm is used for image processing; otherwise, the inter-frame difference method is used for image processing. Image processing, and finally perform morphological filling processing on the image processed by the inter-frame difference method or the image processed by the Vibe algorithm.

Preferably, the pixel-level image stability judgment includes comparing the pixel value of each pixel in the current frame with the pixel value of the pixel at the same position in the previous frame, and if the number of changed pixels is less than the pixel identification threshold , it is judged as a stable state, otherwise it is judged as an unstable state.

Preferably, the inter-frame difference method includes performing difference between two consecutive frames of images in the image sequence to obtain a grayscale difference image, then binarizing the grayscale difference image to extract motion information, and detecting and segmenting the image obtained by inter-frame change regions The background area and the moving object area are distinguished, and then the moving target to be detected is extracted.

Preferably, image filtering is performed on the image processed by the inter-frame difference method, specifically using the neighborhood method, that is, the pixels in the neighborhood in the image processed by the inter-frame difference method are scanned one by one, and each pixel in the neighborhood of the pixel point is The pixel values of the points are sorted from small to large or from large to small, and the intermediate value of the pixel value of each pixel is obtained, and the intermediate value is assigned to the pixel corresponding to the current point in the image processed by the inter-frame difference method.

Preferably, the neighborhood is a circular area with the current pixel as the center and the length of 3 pixels as the radius.

Preferably, image filtering is performed on the image processed by the inter-frame difference method, specifically using the two-dimensional sliding template method, that is, the pixels in the two-dimensional template W are sorted according to the size of the pixel value, and a monotonically rising or falling two-dimensional data sequence is generated. And calculate the median, two-dimensional median filter output g(x,y)=med{f(x-k,y-l),(k,l∈W)}, where, f(x,y), g(x,y ) are original image pixels and processed image pixels respectively, W is a two-dimensional template, x, y represent the coordinate values of pixel points, k, l represent the step value in the two-dimensional template, med{} represents the pair The median value after sorting the pixel values of the pixels in the 2D template.

Preferably, the morphological filling process includes an open operation, that is, the process of first corroding and then expanding;

The erosion includes scanning each pixel in the image with a structural element, and performing an "AND" operation with each pixel in the structural element and the pixels covered by it. If both are 1, the pixel is 1, otherwise it is 0;

The dilation includes using a structural element to scan each pixel in the image, and performing an AND operation with each pixel in the structural element and the pixels it covers. If both are 0, the pixel is 0, otherwise it is 1.

Preferably, the morphological filling process includes a closed operation, that is, the process of dilation first and then erosion;

The erosion includes scanning each pixel in the image with a structural element, and performing an "AND" operation with each pixel in the structural element and the pixels covered by it. If both are 1, the pixel is 1, otherwise it is 0;

The dilation includes using a structural element to scan each pixel in the image, and performing an AND operation with each pixel in the structural element and the pixels it covers. If both are 0, the pixel is 0, otherwise it is 1.

Preferably, it is judged whether to end the collection, if the collection is not finished, continue to collect, and save the collected data in the buffer, repeat the whole process; if the collection is over, save the data in the collection buffer to the memory , and release the buffer memory, and end the operation.

Compared with the prior art, the present invention uses the inter-frame difference method and morphological filling to improve the defects of the Vibe algorithm, and greatly improves the detection accuracy of moving objects in the image.

Description of drawings

Fig. 1 is the flow chart of the first preferred embodiment of the method for improving the detection accuracy of moving objects in images according to the present invention;

Fig. 2 is a flow chart of the second preferred embodiment of the method for improving the detection accuracy of moving objects in images according to the present invention;

Fig. 3 is the image contrast chart after original image and inter-frame difference method processing;

Figure 4 is a comparison of the original image and the image after morphological filling.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings.

As shown in Fig. 1, the present invention first carries out image acquisition, then carries out pixel-level image stability judgment to each frame image of acquisition, if stable, then adopt Vibe algorithm to carry out image processing, otherwise, adopt inter-frame difference method to carry out Image processing, and finally perform morphological filling processing on the image processed by the inter-frame difference method or the image processed by the Vibe algorithm, so as to eliminate the hole phenomenon generated in the motion detection process, and improve the detection accuracy of moving objects through the above methods .

For image collection, the present invention can store the collected images into the internal memory of the computer through the camera. During implementation, the camera can be directly opened for image acquisition by using a camera-driven application programming interface (Application Programming Interface, API for short) provided by the windows system. For example, turn on the camera device and set the image acquisition parameters of the camera; when the program opens the camera, it needs to set the parameters for the camera, here is mainly the pixel value of each frame. Since the program runs on the windows system, and the camera device used is driven by the system, it saves the time overhead of writing the camera driver code by yourself, and only needs to use the camera driver API provided by the windows system in the program code interface.

For the pixel-level image stability judgment, the pixel value of each pixel in the current frame is compared with the pixel value of the same position pixel in the previous frame. If the number of changed pixels is less than the pixel recognition threshold, the judgment is a stable state, otherwise it is judged as an unstable state

When repeatedly processing multiple frames of images, when the moving object is detected to be unstable, switch to the inter-frame difference method, and then when the moving object is detected to be stable, re-initialize the background sample set and switch to the Vibe algorithm.

Since the Vibe algorithm is initialized from the first frame of image, at the beginning of initialization, the entire collected image will be the foreground image by default, that is, the entire image is in motion. Therefore, it is necessary to compare the images of the next few frames to restore the scene, so that it can judge the foreground image and the background image of the entire scene, and separate the foreground image and the background image. When a moving object enters the acquisition area, the pixel-level image stability judgment is first performed on the image. If it is unstable, use the inter-frame difference method. If it is stable, switch to the Vibe algorithm. If it is judged to be stable, it indicates that the entire moving object is already in a stable state. At this time, when using the Vibe algorithm to detect the moving image, it will be eliminated. Drag shadow.

For the inter-frame difference method, the region that changes in two adjacent frames of images is detected. The method is to use the difference between two consecutive frames of images in the image sequence to obtain a gray-scale difference image, and then binarize the gray-scale difference image to extract motion information. The image obtained by detecting and segmenting the change area between frames is used to distinguish the background area and The moving object area, and then extract the moving target to be detected.

The inter-frame difference method of the present invention is to compare the gray value of the pixel point corresponding to the two frames of images in the image sequence, and subtract the gray value of the pixel point corresponding to the two frames. If the difference is very small, it can be considered that there is no moving object passing through the point, otherwise If the gray level changes greatly, it is considered that there is an object passing by.

For example, the change between the image f _k (x, y) and f _k+1 (x, y) of the kth frame and the k+1 frame is represented by a binary difference image D(x, y), such as formula (1 ):

T is the difference image binarization threshold, because there is no need to update the background model, so when the moving object is unstable, the inter-frame difference method can be preferably considered.

In the above embodiment, the frame difference method is added to the traditional Vibe algorithm, so as to improve the Vibe algorithm which has the defect of dragging shadows.

Preferably, as an optional method, the present invention performs image filtering on the image processed by the inter-frame difference method, as shown in Figure 2, which can be performed in a variety of ways, including but not limited to the neighborhood method and the two-dimensional sliding template method Wait.

The neighborhood method includes the following processes:

First, get the first address of the source image and the width and height of the image.

Secondly, open up a memory buffer to temporarily store the result image, and initialize it to 0. Then, scan the pixels in the image one by one, sort the elements of its neighborhood from small to large, and assign the obtained intermediate value to the pixel corresponding to the current point in the target image. The neighborhood is a circular area with the current pixel as the center and the length of 3 pixels as the radius.

Then loop the previous step until all the pixels of the source image are processed.

Finally, the result is copied from the memory buffer into the source image's data buffer.

The neighborhood method adopted in this embodiment is a median filtering method based on the spatial domain. The median filtering method is a non-linear smoothing technique, which sets the pixel value of each pixel to the median value of all pixel values within a certain adjacent area of the point. Median filtering is a non-linear signal processing technology that can effectively suppress noise based on sorting statistics theory. The basic principle of median filtering is to use the pixel value of a point in a digital image or digital sequence with each pixel in a neighborhood of the point The median value of the pixel value of the point is replaced, so that the pixel value of the surrounding pixel points is close to the real value, thereby eliminating isolated noise points.

The two-dimensional sliding template method sorts the pixel points in the template according to the size of the pixel value, and generates a monotonically rising (or falling) two-dimensional data sequence. Two-dimensional median filtering output g(x,y)=med{f(x-k,y-l),(k,l∈W)}, where f(x,y) and g(x,y) are the original image Pixels and processed image pixels, W is a two-dimensional template, usually 3×3, 5×5 areas, and can also be different shapes, such as linear, circular, cross-shaped, circular, etc.; x, y represents the coordinate value of the pixel, k and l represent the step value in the two-dimensional template, and med{} represents the intermediate value after sorting the pixel values of the pixel in the two-dimensional template.

The improved Vibe algorithm can detect moving targets, but due to the characteristics of the Vibe algorithm itself and the characteristics of the frame difference method, when using the Vibe algorithm to detect moving objects, the detected targets often appear hollow, although The void is not very obvious, but it is there. Figure 3 shows the original image (left) and the image processed by the inter-frame difference method (right). The outline is very clear from the whole, and the processing effect is good, but there are still holes, especially the head hole. serious. Therefore, it is necessary to further deal with the existing void phenomenon.

The present invention uses a morphological filling method to fill the image processed by the inter-frame difference method or the image processed by the Vibe algorithm or the image processed by filtering to eliminate the cavity phenomenon and further improve the accuracy of detecting moving objects.

Morphological filling uses a "probe" called a structural element to collect image information. When the "probe" is constantly moving in the image, the relationship between various parts of the image can be examined to understand the structural characteristics of the image. Its core idea includes opening operation and closing operation.

The calculation model of the opening operation is:

Dst1=open(src,element)=dilate(erode(src,element)) (2)

In the formula (2), open() represents the open operation, src represents the original image, element represents the processed image, dilate() represents the erosion operation, and erode() represents the expansion operation

The opening operation is a process of erosion first and then dilation. Used to eliminate small objects, separate objects at thin points, and smooth the boundaries of larger objects without significantly changing their area. The opening operation is usually used when it is necessary to remove small particle noise and disconnect the connection between objects, and has the advantage of basically keeping the original size of the object unchanged.

The calculation model of the closing operation is:

Dst2=close(src,element)=dilate(erode(src,element)) (3)

Among them, close() in formula (3) represents the closing operation, src represents the original image, element represents the processed image, dilate() represents the erosion operation, and erode() represents the expansion operation

The closing operation is a process of first dilation and then erosion, which can fill the small holes in the object and smooth the object boundary.

As a preferred method, the opening operation can be performed first and then the closing operation is performed. After each frame of image is opened and filled by the closing operation, the holes of moving objects in the entire picture will be filled.

The function of the erosion is to eliminate the boundary points of the object, shrink the target, and eliminate the noise points smaller than the structural elements. The specific operation of corrosion is: scan each pixel in the image with a structural element (generally 3×3 in size), and perform an "AND" operation with each pixel in the structural element and the pixels covered by it. If both are 1, Then the pixel is 1, otherwise it is 0.

The effect of the expansion is to merge all the background points in contact with the object into the object, so as to increase the size of the target and fill the holes in the target. The specific operation of expansion is: scan each pixel in the image with a structural element (generally 3×3 in size), and use each pixel in the structural element and the pixels covered by it to perform an "AND" operation. If both are 0, Then the pixel is 0, otherwise it is 1.

After the picture is processed by morphological filling, the hollow phenomenon of the whole image can be well filled, so that after the whole picture is processed by the invention, moving objects can be well displayed. As shown in Figure 4, the original image (left) and the image after morphological filling processing (right), the outline of the entire moving object is very clear, and the cavity phenomenon is much reduced compared with that before processing, and the improvement effect is obvious.

Finally, judge whether the collection is over, if the collection is not over, continue to collect, save the collected data in the buffer, and repeat the above operation; if the collection is over, turn off the device and save the data in the collection buffer The data is saved to the memory, and the buffer memory is released, and the operation ends.

The invention adopts the inter-frame difference method and morphological filling to improve the defects of the Vibe algorithm, and can improve the real-time detection accuracy of the moving object in the image.

The above examples have further described the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above examples are only preferred implementations of the present invention and are not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made to the present invention within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (9)

1. A method for improving the detection accuracy of moving objects in an image, characterized in that: each frame image collected is judged for pixel-level image stability, if it is stable, the Vibe algorithm is used to process the image, otherwise, the frame is used The frame-to-frame difference method is used to process the image, and finally the image processed by the frame-to-frame difference method or the image processed by the Vibe algorithm is processed by morphological filling. 2. The method for improving the detection accuracy of moving objects in images according to claim 1, characterized in that: said pixel-level image stability judgment includes that the pixel values of each pixel in the current frame are at the same position as the pixel in the previous frame The pixel values are compared, if the number of changed pixels is less than the pixel recognition threshold, it is judged as a stable state, otherwise it is judged as an unstable state. 3. The method for improving the detection accuracy of moving objects in images according to claim 1, characterized in that: the inter-frame difference method comprises using two consecutive frames of images in the image sequence to perform a difference to obtain a gray-scale difference map, and then binarize The motion information is extracted from the gray difference image, and the background area and the moving object area are distinguished from the image obtained by detecting and segmenting the changing area between frames, and then the moving object to be detected is extracted. 4. The method for improving the detection accuracy of moving objects in images according to claim 1, characterized in that: image filtering is performed on the image processed by the inter-frame difference method, specifically using the neighborhood method, that is, scanning one by one and processing by the inter-frame difference method After the pixels in the neighborhood in the image, the pixel values of each pixel in the neighborhood of the pixel are sorted from small to large or from large to small, and the median value of the pixel value of each pixel is obtained, and the median Assign the value to the pixel corresponding to the current point in the image processed by the inter-frame difference method. 5. The method for improving the detection accuracy of moving objects in images according to claim 4, wherein the neighborhood is a circular area with the current pixel as the center and the length of 3 pixels as the radius. 6. The method for improving the detection accuracy of a moving object in an image according to claim 1 is characterized in that: image filtering is performed on the image processed by the inter-frame difference method, specifically using a two-dimensional sliding template method, that is, within the two-dimensional template W The pixels are sorted according to the size of the pixel value, generate a monotonically rising or falling two-dimensional data sequence and calculate the median value, and the two-dimensional median filter output g(x,y)=med{f(x-k,y-l),(k,l ∈W)}, where f(x,y) and g(x,y) are the original image pixel and the processed image pixel respectively, W is the two-dimensional template, x and y represent the coordinate value of the pixel, k , l represents the step value in the two-dimensional template, and med{} represents the intermediate value after sorting the pixel values of the pixels in the two-dimensional template. 7. The method for improving the detection accuracy of a moving object in an image according to claim 1, wherein: the morphological filling process includes an opening operation, that is, the process of first corroding and then expanding; The erosion includes scanning each pixel in the image with a structural element, and performing an "AND" operation with each pixel in the structural element and the pixels covered by it. If both are 1, the pixel is 1, otherwise it is 0; The dilation includes using a structural element to scan each pixel in the image, and performing an AND operation with each pixel in the structural element and the pixels it covers. If both are 0, the pixel is 0, otherwise it is 1. 8. The method for improving the detection accuracy of a moving object in an image according to claim 1, wherein: the morphological filling process includes a closed operation, that is, the process of first expanding and then corroding; The erosion includes scanning each pixel in the image with a structural element, and performing an "AND" operation with each pixel in the structural element and the pixels covered by it. If both are 1, the pixel is 1, otherwise it is 0; The dilation includes using a structural element to scan each pixel in the image, and performing an AND operation with each pixel in the structural element and the pixels it covers. If both are 0, the pixel is 0, otherwise it is 1. 9. The method for improving the detection accuracy of moving objects in images according to claim 1, characterized in that: judge whether to end the collection, if the collection is not finished, continue to collect, and save the collected data in the buffer, repeat The whole process; if the acquisition ends, save the data in the acquisition buffer to the memory, release the buffer memory, and end the operation.