Title:Introduction of a tree-based technique for efficient and real-time label retrieval in the object tracking system

Abstract:This paper addresses the issue of the real-time tracking quality of moving objects in large-scale video surveillance systems. During the tracking process, the system assigns an identifier or label to each tracked object to distinguish it from other objects. In such a mission, it is essential to keep this identifier for the same objects, whatever the area, the time of their appearance, or the detecting camera. This is to conserve as much information about the tracking object as possible, decrease the number of ID switching (ID-Sw), and increase the quality of object tracking. To accomplish object labeling, a massive amount of data collected by the cameras must be searched to retrieve the most similar (nearest neighbor) object identifier. Although this task is simple, it becomes very complex in large-scale video surveillance networks, where the data becomes very large. In this case, the label retrieval time increases significantly with this increase, which negatively affects the performance of the real-time tracking system. To avoid such problems, we propose a new solution to automatically label multiple objects for efficient real-time tracking using the indexing mechanism. This mechanism organizes the metadata of the objects extracted during the detection and tracking phase in an Adaptive BCCF-tree. The main advantage of this structure is: its ability to index massive metadata generated by multi-cameras, its logarithmic search complexity, which implicitly reduces the search response time, and its quality of research results, which ensure coherent labeling of the tracked objects. The system load is distributed through a new Internet of Video Things infrastructure-based architecture to improve data processing and real-time object tracking performance. The experimental evaluation was conducted on a publicly available dataset generated by multi-camera containing different crowd activities.