Research On Real-time Detection Of Moving Target Under Shifting Background

Moving object detection can be used as a basic step for many visual applications,the purpose of detection is to separatethe moving objects(foreground)from the static background.Under the assumption of a fixed camera,there are many related algorithms.This assumption constrains the image pixels that correspond to the background to maintain their positions in consecutive frames and the scope of application of these algorithms.This paperstudies the detection of moving objects with a moving camera.The main difficulty in moving target detection under shifting background is that the target pixel and the background pixel are displaced in the image.Second,real-time is also a need to pay attention to the problem.In this paper,the motion compensation technique is used to transform the target detection problem undershifting background into the target detection problem under the static background.Finally,the background modeling method is used to detect the moving target.In the process of estimating the motion model,this paper designs a sparse background motion extraction method,which can eliminate the interference of foreground motion better,and extract the background motion distribution more evenly.Finally,using RANSAC and least squares method,The robustness of the motion model parameters should be estimated.In the background modeling stage,the dual-mode single-Gaussian model is used as the background model,and the probabilistic map and the sampling mask are used to improve the detection effect of the dual-mode single Gaussian model.On the one hand,the sampling mask is used to differently update the different regions Strategy and the choice of the prospective decision to reduce the false test rate,on the other hand the use of foreground probability to construct adaptive decision threshold to improve the fixed threshold defects.Comparisons with other algorithms on a unified data set,the experiment results show that the proposed algorithm has a good performance in the detect results,and the time per frame can satisfy the real-time requirements.To further improve the performance of the algorithm,we will try to integrate the motion information of the pixel in the background modeling process.