Research On Target Extraction Algorithm By ViBe Model And Shadow Elimination

In recent years,with the rapid development of computer technology and digital vision,intelligent video surveillance has been playing an increasingly important role in facilitating the modern life.The moving target extraction technology is an essential fraction in intelligent monitoring systems and has become a research hotspot in computer vision technology.Meanwhile,it is complicated to improve the accuracy of extraction and while adapting different background environments.The commonly used algorithms for moving target extraction were studied in this thesis.Focusing on the ghost and shadow problems,the model of visual background extraction(ViBe)was thoroughly studied for moving target extraction system and corresponding improvements were described in details.Firstly,the suppression and elimination of ghost and the adaptive threshold algorithm were proposed in this thesis.Secondly,the algorithm combined with spatiotemporal scale invariant local ternary pattern(ST-SILTP)texture features to achieve shadow extraction and elimination was proposed.Then the moving target extraction system with improved ViBe model was designed and implemented,and the practicability of the system was verified by a series of experiments.The main research work is as follows:1.For the algorithms of moving target extraction,the optical flow method,frame difference method and Gaussian background model,Codebook background model and ViBe model in background subtraction were studied in this thesis.And the advantages and disadvantages of these algorithms were analyzed.2.ViBe model was easy to introduce ghost and extracted a lot of false foreground in complex dynamic background.In order to solve these problems,the suppression and elimination of ghost and the self-adaptive threshold algorithms were proposed.For the problem of ghost,the improved initialization and neighborhood pixel space consistency principle were combined to eliminate ghost.Considering a lot of false extraction in complex dynamic background,the information of the background sample-set and the neighborhood information of the current pixel were combined in the thesis to reflect the background complexity.That could be self-adaptive segmentation threshold.The improved algorithms could achieve the suppression and elimination of ghost and improve the accuracy of extraction.3.The shadow of moving target was also extracted as the foreground due to misjudgment of shadow of moving targets in ViBe model,which might reduce accuracy of the extractions.To overcome this issue,the algorithm of combining the ViBe model with the ST-SILTP texture features was proposed after studying the nature of the moving shadow and the texture features.Besides,experimental results showed that the improved method could solve the shadow problem in developing the effect of target extraction.4.The moving target extraction system with improved ViBe model was designed and realized in this thesis.The system could not only extract motion target for the recorded video,but also extract the moving target for the real-time video.