Adaptive Clustering Algorithm Based Small Group Detection And Tracking

The detection and tracking of small groups is the key technology of intelligent video surveillance system,and the basics of high-level vision tasks,such as abnormal event detection,behavior understanding,scene understanding and so on.Small group refers to several action consistency groups in the moving region of close proximity.Small groups in video surveillance not only reflects the society behavior and safety problems,widely used,but it is very challenging in the field of computer vision.The detection and tracking of small groups depends on the detection and tracking of individual moving targets,also depends on acquiring algorithm description and group characteristics.This involves many fields of image processing,pattern recognition and machine learning knowledge,so it has high value of theoretical research.In recent years,many analysis algorithms of groups have been proposed,but due to the activities distribution of crowd motion,structure dynamic,small group changes frequently in video surveillance,mutual occlusion and complex background problems for small group analysis has brought quite a challenge.Therefore,this paper studies a method of detection and tracking of small groups based on the adaptive clustering method,by tracking the improved multi-objective optimization,measurement of trajectory similarity,and adaptive clustering,to find and track small groups.The main work of this paper is divided as two points:(1)The algorithm of tracking individual of the moving target is proposed based on the fitting of the pre-estimated trajectory of two-way velocity.It is aimed at problems about people may appear to obscure each other and affect by complex background in surveillance video.The purpose is for providing more accurate input data for the subsequent examination of the detection and tracking of small groups based on adaptive clustering algorithm.In public set of FM_dataset and SMOT,experimental results show that the temporal trajectory track fitting method based on bidirectional speed pre sentenced is put forward to restore a target when short-time occlusion,greatly improving the accuracy of target tracking.(2)The paper proposes a clustering algorithm of small groups based on adaptive detection and tracking algorithm.It is based on multiple object detection and tracking trajectory of space and time,through the measure of similarity,based on the measured scene histogram statistical distribution,the determination of the adaptive segmentation of small groups threshold,and then through the clustering algorithm of adaptive judgment small groups of splitting and merging.Based on the evaluation index,such as accuracy,fault rate,misdiagnosis rate,the results on FM_dataset datasets verify the robustness of the proposed method.