Multiple human tracking using 3D ellipsoidal model and SIR-Particle Filter technique

Visual tracking is a fundamental key to the recognition and analysis of human behavior. This paper presents an approach to track several humans from video sequences acquired in real time. It addresses the key concerns of real time performance and continuity of tracking in overlapping and non-overlapping fields of view. It represents the human body by a parametric ellipsoid in a 3D world. The elliptical boundary can be projected rapidly, several hundred times per frame, onto any image for comparison with image data within likelihood mode. This is implemented by using SIR-Particle filter for tracking multiple humans. Adding variables to encode visibility and persistence into the state vector, it tackles the problem of distraction and short period occlusion. The accuracy of the algorithm is evaluated using the metric, Multiple Object Tracking Accuracy (MOTA) and found that the accuracy was increased up to 90%, when tested with the benchmark dataset.