Research On Object Detection And Tracking Method Based On Improved FBS-ABL Background Modeling And Particle Filtering

Moving target detection and tracking as an important branch of computer vision,in intelligent video surveillance,industrial robots,UAVs and other fields have been widely used,in practical applications,the robustness of the algorithm will be affected by the dynamic environment or target scale changes,the current complex environment of moving target detection and tracking is still a challenging work,this paper based on background subtraction,combined with the research results of predecessors,the shortcomings of the existing algorithm are improved,the main work is as follows:(1)In this paper,four background modeling methods of GMM,PBAS,Vibe and FBS-ABL are first studied,and the FBS-ABL algorithm is determined as the research method of this paper through experimental comparison,aiming at the problem of "ghosting" of FBS-ABL algorithm in the multimodal complex background such as falling leaves,lighting and roads,a moving target detection algorithm based on color histogram is proposed,and the color characteristics of the target and its target neighborhood are compared for the segmented foreground target.The Bhattacharyya coefficient is used to match the feature similarity,and whether the target region is "ghosted" is judged according to the similarity,and for the "ghost" region,reset its gray value and update the background model of the region to suppress the "ghost" generated by subsequent frames.The effectiveness of the algorithm is proved by experimental verification of public datasets in different scenarios.(2)Aiming at the target loss and drift caused by target occlusion,lighting change and other factors,an improved particle filter target tracking algorithm is proposed,the algorithm combines FBS-ABL and color particle filtering,color particle filtering is introduced into the image local entropy concept to describe the target model,so that it has good robustness when it is in the lighting environment,the FBS-ABL algorithm can accurately locate the moving target area,avoid the error caused by manual selection,when the target is occluded and the scale changes,The target can be quickly located to prevent the loss of the target,and the experimental results show that the algorithm performs well in target tracking under illumination and occlusion conditions.