Research On Moving Object Detection Method Based On Online Robust Principal Component Analysis

With the explosive development of artificial intelligence technology,the research in the related fields has also begun to be paid more attention by researchers all over the world.Since the information acquired by vision is 80%of the proportion of all human perceiving organs,the development of artificial intelligence is obviously crucial before the development of information and the final stage of decision-making.The detection of moving object,as an important process of computer vision,has a wide range of applications.However,existing algorithms for moving object detection still have many problems such as low robustness,low detection accuracy and poor real-time performance,so they have great research significance and room for improvement.In general,the main work and innovation of this paper can be summarized as the following parts:1.The traditional object detection algorithms are roughly divided into the following major categories:temporal difference,background subtraction,optical flow method and moving object detection based on robust principal component analysis(RPCA).After an overview of their principles,the advantages and disadvantages of these algorithms are summarized,although some of them have simple principle,fast calculation speed,and get the result of detection accuracy in certain scenes,but in general,they cannot meet the needs of practical application.2.The origin and theory of robust principal component analysis model is introduced,and several excellent algorithms for solving RPCA models are discussed.Three common problems are summarized:The computation of singular value decomposition step is large,it is not suitable for scenes with high real-time requirement,and the sample occupies large memory.3.Aiming at the poor initialization effect caused by random initialization strategy in online robust principal component analysis(OR-PCA),this paper proposes an object detection algorithm combined with batch initialization,named improved OR-PCA.By the batch initialization strategy,the initialization efficiency is improved and the size of the base matrix can be reduced to improve the efficiency of detection.A series of experimental results show that the improved OR-PCA algorithm proposed in this paper has better detection accuracy,computation speed and robustness.Compared with the original OR-PCA algorithm,the initialization efficiency is greatly improved.4.By introducing the idea of superpixel segmentation and further utilizing the color and spatial structure information in limited initialized video frames,we propose an improved OR-PCA moving object detection algorithm based on superpixels,which improves the algorithm's anti-jamming performance for high dynamic background scenes.Finally,the median filter is selected as the post processing,and a holonomic moving object detection software is completed,which is displayed in the form of user interface.