| Subject of this article is supported by Project Things of WuXi City Management.Video surveillance in public safety management and security plays an important role, especially the entrance and exit in the subways, airports and other public transport sites, not only saving labor costs for corporate, more in the 24-hour achieve real-time monitoring of early warning function. There are the problems of high rate of false positives and low efficiency in current crowd detection algorithm. Thus, in the crowded video surveillance detection algorithm to improve the accuracy and stability is necessary. This is currently a hot research field of computer vision.In this paper, the entropy gradient-based congestion detection algorithm based on analysis of the test, focusing on the goal of moving target detection and classification techniques, the introduction of these two technologies to improve the algorithm research.This paper outlines the goal of moving target detection and classification, and clustering-based modeling to improve the ideological background of the program are summarized. In the analysis of SVM classification algorithm, based on HOG features and support vector machines for face detection algorithm, based on the entropy gradient detection algorithm based on congestion, we propose a target recognition based on improved adaptive threshold algorithm using SVM algorithm for target classification, the classification results for "character" of the video frame to determine the congestion, while real-time updates by updating the background threshold, threshold with changes in the environment to achieve adaptive. Based on this algorithm designed and implemented a prototype system to monitor congestion detection, test results indicate that the system under normal conditions, crowded with good detection capability in the environment of rapid change, complex background of the case, the system can still achieve good congestion detection warning effect. |
PDF Full Text Request