Multi-Level Clustering Algorithm for Pedestrian Trajectory Flow Considering Multi-Camera Information

With the development of surveillance video systems from a single camera to a multi-camera network, the traditional video target trajectory clustering algorithm does not consider the camera position and field of view, resulting in poor clustering results. To solve this problem, this paper proposes a multi-level clustering algorithm for semi-supervised video target geographic flow. The proposed method is based on flow space and uses geographic flow hierarchical representation for video target trajectories to obtain geographic flow sub-segment sets at different levels. Meanwhile, this method takes the camera number as a label and performs semi-supervised clustering on the flow sub-segment sets at different levels. In this way, the clustering center set at the corresponding level is obtained, and the general trend of a large number of geographic flows is effectively described with cluster centers. The proposed method makes up for the shortcomings that traditional algorithms can only cluster at a single level and do not consider camera information, and its effectiveness is verified by a series of experiments.