Abnormal Event Detection In Video Sequences Based On Low Rank Approximation

In recent years,the low-rank approximation theory has been widely used in pattern recognition and computer vision.For example,collaborative filtering,image registration,video denoising and other research,but it is rarely applied for abnormal events detection in video sequences.While considering the low-rank property of video data,this paper makes improvements for the low rank approximation algorithm theory,and applies it to solve the problem of abnormal events detection in video sequences.The works of the article are as follows:(1)The method of abnormal events detection in video sequences based on the low-rank approximation structured sparse representation is proposed.The key problems include how to discover the key information of behavior patterns and abandon the redundant contents from the huge video data,and how to improve the efficiency of abnormal events detection.To solve these problems,the low-rank structured sparse coding model is introduced in the paper,and the method of abnormal events detection in video sequences based on the low-rank approximation structured sparse representation is proposed.Experimental results on the public datasets are shown that the method can improve the accuracy of abnormal events detection and also greatly increase the time efficiency.(2)The method of an adaptive anomaly detection based on low rank structured sparse representation is proposed.The method of abnormal events detection in video sequences based on the low-rank approximation structured sparse representation sometimes detects normal test samples as abnormal events,resulting in the high false positive rate of abnormal event detection.Considering the low-rank information in dictionary learning has not been fully exploited,this paper combines the low-rank structured sparse dictionaries and their corresponding low-rank information for video events reconstruction,and presents an adaptive anomaly detection method based on low rank structured sparse representation.The method contains two main stages,which are the dictionary learning based on low-rank structured sparse representation and the weighted reconstruction based on low-rank sparse dictionary.Experimental results demonstrate that the approach can effectively improve the detection accuracy while ensuring the high time efficiency.(3)An improved adaptive anomaly event detection method based on low-rank structured sparse representation is implemented.Aiming at the low time efficiency of directly utilizing the low-rank structured sparse coding model for dictionary learning,this paper takes the low-rank structured sparse dictionary as the initial dictionaries,further fitting the target function,and implements an improved adaptive anomaly event detection method based on low-rank structured sparse representation.Our experiments reveal that the method can enhance the time efficiency while guaranteeing the detection accuracy.