Research And Optimization Of Video's Moving Object Detection Algorithm

As an important part of computer image processing technology,the application of moving object detection in real life is becoming more and more widespread.Not only the security monitoring of public places or the behavior recognition of artificial intelligence,but also it can even be seen in the military field.For different scene changes and problems encountered in the detection process,scholars of various research institutions and institutions have successively developed a variety of algorithms,but for the existing algorithms,there is no real-time,efficient and robust algorithm to adapt to all environments.In this thesis,the problem of misdetection of robust principal component analysis algorithm in dynamic context and the problem of missed detection due to slow motion are studied in depth.The main work is as follows:The thesis first studies the principle of the three subspace algorithms,principal component analysis algorithm,robust principal component analysis algorithm and two-level robust principal component analysis algorithm,and discusses and analyzes the advantages and disadvantages of these three algorithms.Aiming at the problem that the robust principal component analysis algorithm can not deal with the complicated dynamic background such as shaking leaves,water surface fluctuations and fountains,this paper proposes an improved algorithm.The Gaussian function is used to fit the dynamic background points in the time domain to segment the dynamic background points and moving target points detected by the robust principal component analysis algorithm.Differences of the mean and variance of the dynamic background points and moving target points in the time domain is also used.In order to improve the robustness of the algorithm,this paper particularly proposes to use block density instead of pixels to perform operations.The experimental results show that the dynamic background problem in the detection results of the robust principal component analysis algorithm is basically solved.In order to solve the problem of missed detection of robust principal component analysis algorithm when dealing with slow moving targets and even stationary moving targets,the thesis proposes to use saliency detection algorithms to optimize the processing of the robust principal component analysis algorithms.The dissertation divides the video scene into simple and complex scenes for processing.For simple scenes,the paper superimposes the results of saliency detection with the results of moving target detection.For complex scenes,the paper first filters the staticallysignificant targets in the saliency detection results,and then extracts the motion in the moving target detection results.Using this region's saliency test results to optimize the processing of voids inside moving objects.The experimental results show that after the saliency detection algorithm is combined,the problem of missed detection of the robust principal component analysis algorithm is well solved.