Research On Background Subtraction Algorithm Based On Low Rank Sparse Decomposition

Moving object detection in the video has always been a hot research direction in computer vision.Although the traditional methods have achieved some results in target detection,the foreground detection in the complex environment still needs to be improved.At present,there are many achievements in background subtraction,such as robust principal component analysis and robust principal component analysis based on tensors.In order to cope with the challenges brought by more complex situations,we continue to study background subtraction based on robust principal component analysis in this paper,and the research findings are as follows:（1）An algorithm via l_1/2norm and saliency constraint is proposed.In the new method,the l_1/2norm is applied to constrain the foreground to improve the target detection accuracy in dynamic background.The saliency constraint is introduced to detect slow-moving targets.Experimental results show that the proposed algorithm achieves the optimal comprehensive index value compared with the six state-of-the-art methods under dynamic background disturbance.Subjectively,the proposed algorithm can significantly improve the accuracy of foreground detection,especially when the target is moving slowly.（2）Focus on the problem of large area of dynamic background and bad weather,we propose a moving target detection model via tensorγnorm and G（2,1）norm.In the model,the tensorγnorm,which is extended over the matrixγnorm,is applied to approximate the background.The G（2,1）norm,which is proposed in this paper,and the total variation is implemented to constrain the foreground to reduce the interference of the dynamic background.Compared with the recently advanced algorithms,experiments show that the proposed method performs better numerically and visually,especially in the dynamic background and severe weather.