Research On Moving Target Detection Via Optimized Low Rank Matrix Algorithm

With the rapid development of science and technology,intelligent technologies such as intelligent transportation related to machine vision,intelligent security,and smart homes developed rapidly in recent years have widely affected people's lives.As one of the key technologies in the field of machine vision,target detection has always attracted considerable interests.Usually,object detection is to extract the relatively moving targets in video sequences,and the results directly affect the quality of subsequent analysis,recognition and processing.The accurate extraction of the moving object region is always the focus of video moving target detection.This thesis focuses on the background modeling method based on low-rank matrix decomposition,develops video moving object detection research,proposes an optimization low-rank matrix background modeling method with time continuity constraints,and applies it to video motion target detection to improve the accuracy of video motion targets;in addition to the shadow problem of moving targets commonly found in practical applications,a method of target shadow detection using fusion textures and color features is proposed to effectively remove the influence of cast shadows on the motion target extraction and further achieve high overall accuracy of moving target detection.The main work of this thesis is mainly reflected in the following three aspect:1.Learning common moving object detection methods,including inter-frame difference method,optical flow method and background subtraction method,and numerically implementing and comparing several representative methods.The background modeling method of low-rank matrix decomposition in background subtraction method is particularly studied in depth.It is pointed out that the importance of time continuity constraint in low rank matrix decomposition model applied to video moving object detection.2.An optimized low-rank matrix decomposition moving target detection method is proposed.This method considers the temporal continuity of the moving target in video sequences,and introduces the time continuity constraint into the low-rank matrix decomposition background modeling model.The optimization model can effectively overcome the disadvantages of the low-rank matrix decomposition of the regular parameters or the segmentation threshold in the background model caused by the improper setting of the moving object extraction accuracy.Experimental results show that this optimization method has two main advantages:First,it can effectively remove the non-continuity interference introduced by imaging equipment or imaging process such as noise.The second is that it can overcome the foreground imaginary shadow in the background model when extracting a large moving target.3.For the problem of cast shadows,an improved N-LBP texture feature operator which introduces neighborhood frame constraints is proposed,and the texture features are fused with color features to remove the target shadow.Experimental results show that the effective shadow removal method is necessary to improve the accuracy of video moving target detection.