The Research Of Crowd Clutters Based On Coherent Filter

Smart city is an important carrier of highly intensive social cluster behavior, and when the scale of the city is larger, the function is the more complex, and the problems arise more, and the potential crisis greater.Scientific planning and management of large scale cities need to be supported by effective social cluster analysis.If the cluster behavior analysis method is ineffective in the management of large-scale events, large-scale assembly and large-scale folk activities, it is likely to lead to disastrous consequences.For example, "2014.12- 31" Shanghai Bund crowded stampede, one of the important reasons is the failed to cluster large-scale acts of perception and make timely judgments.The analysis of video data in urban information network system is an important approach to the analysis of cluster behavior.In recent years, the analysis of the cluster behavior based on video in the field of computer vision has been developed vigorously.At present, the research direction of the cluster behavior analysis mainly include the following aspects: 1 mining analysis of the global motion state; 2 clustering behavior region semantic analysis; 3 clustering behavior anomaly detection; 4 cluster behavior tracking detection and group classification.In a large number of cluster behavior analysis algorithm, the cluster behavior analysis based on agent model is widely used.Agent model is a combination of computer vision, psychology, sociology, game theory and other disciplines, and it is proposed that there are two main methods: macro analysis and micro cluster analysis methods.The former focuses on the model and analysis of the cluster, and the latter pays attention to the relationship between the individual.As an agent model based clustering behavior analysis algorithm, Coherent Filtering can be better to merge two kinds of analysis methods of Agent Model.It is not only the global modeling of the cluster behavior, but also the analysis of the relationship among the individuals.But it also has its own limitations, as a kind of universal strong algorithm, it is difficult to combine the environment to do more detailed cluster analysis, and it is need to optimize its core algorithm further or the introduction of other algorithms to assist analysis.In this paper, we mainly study the clustering behavior tracking and group classification based on the Coherent Filtering. The main work is as follows:1.in-depth study of the cluster behavior analysis algorithm based on the improved Coherent Filtering, different from the original Coherent Filtering, computing cluster individual similarity only consider the speed direction.An improved algorithm is proposed to improve the performance of the cluster. The formula for computing the similarity of clusters is optimized. The improved algorithm is better than the previous algorithm in clustering behavior tracking and group classification.2.Based on the analysis of the cluster behavior of the improved Coherent Filtering, an population density estimation algorithm is proposed,which is based on the improved Coherent Filtering. According to the experimental results, the crowd density estimation algorithm based on the improved Coherent Filtering can get better recognition effect, which can accurately estimate the number of clusters.