Enhanced Multiple-Object Tracking in Video Using Hierarchical Kernel-Based Localization and Directional Histograms

This paper proposes an improved method for multiple-object tracking in video using hierarchical kernel-based localization. First, all moving objects (e.g., hands, bodies) are detected in each frame. Next, object direction is determined by comparing histograms and distances with the previous frame, matching objects to kernels with similar characteristics. The method separates overlapping motion paths through kernel labeling. Performance metrics, including multiple objects tracking accuracy (MOTA) and precision (MOTP), validate the approach, with simulations showing faster and more accurate tracking than comparable methods.