Based On The Improved Sparse Reconstruction Algorithm For Pedestrian Anomalies For Analysis And Analysis

Human abnormal behavior analysis is an important research direction in the field of intelligent video surveillance, by using computer vision technology to analysis human behavior to achieve the objective of the intelligent detection. so that we can filter out a lot of useless information, saving a large amount of manpower, and solved the nachtraglichkeit. In this thesis, we concentrate our attention on target tracking and human behavior analysis.At present, most of the tracking algorithm has many shortcomings, such as, slow tracking speed, weak robustness, large amount of calculation, and error tracking, etc. In this thesis, we effectively combine Kalman filter and the Spatial-Temporal Context algorithm together, which can solve the problem of slow tracking speed and error tracking. The Spatial-Temporal Context algorithm estimates the best target location by learning the local context model and the confidence map. The characteristics of the algorithm is that it is for single background and single target tracking, meanwhile since the model learning and target detection is achieved by Fourier change, so the speed of the algorithm is fast. However, this method can only solve the problem of partial occlusion. In order to solve these problems of this algorithm, we use Kalman filter to predict the movement trend of the target, which can deal with shelter problem and improve tracking accuracy.In terms of abnormal analysis, the method of using trajectory to analyze the target behavior is getting more and more attention. Trajectory analysis which based on the model has the problem of low speed, because it has to analyze the details information of the trajectory. Trajectory analysis which based on the high-level intentionality characteristics has the problem of high complexity. In this paper we analyze the trajectory by optimized Sparse Reconstruction Algorithm and distinguish normal or abnormal according to the reconstruction residual. The advantage of this method is that it does not need to analyze the details characteristics of the trajectory. We only need to collect the pedestrian trajectory for subsequent anomaly identification. The experimental results show that the proposed method has higher recognition rate on small sample set.