Automatic labeling of tracked objects based on an indexing mechanism

Real-time object tracking is still a critical challenge in artificial vision research. In such a mission, it is essential to assign a unique identifier or label to each tracked object, regardless of the area, time of appearance, or detector camera, to distinguish it from other objects and to conserve as much information as possible about the tracked objects with the same label. This conservation is a significant issue, especially in largescale video surveillance systems, due to the linear complexity of the sequential search to find the labels of detected objects in data increasing with time, the number of tracked objects, and the number of active cameras in the network. To overcome this problem, we propose a new automatic multi-object labeling solution for efficient real-time tracking based on an indexing mechanism. This mechanism organizes the massive metadata of objects extracted during tracking into a tree-based indexing structure. The main advantage of this structure in a tracking system is 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. This paper discusses the effectiveness of the label search algorithms and the tracking quality compared to other recent tracking systems on real-world datasets. Experimental results showed good performance in reducing search time and improving tracking quality.