Integrated Detection and Tracking for Multiple Moving Objects using Data-Driven MCMC Data Association

Abstract

We propose a framework to address the multiple target tracking problem, which is to recover trajectories of targets of interest over time from noisy observations. Due to occlusions by targets and static objects, parallax or other moving objects, foreground regions cannot represents targets faithfully although motion segmentation is usually computationally efficient. We adopt the real Adaboost classifier to generate meaningful candidate rectangles to interpret the foreground regions. Tracks are generated from these candidates according to the smoothness of motion, appearance and model likelihood overtime. To avoid enumerating all possible joint associations, we take a Data Driven Markov Chain Monte Carlo (DD-MCMC) approach which samples the solution space efficiently. The sampling is driven by an informed proposal scheme controlled by a joint probability model combining motion, appearance and model information. Comparative experiments with quantitative evaluations are provided.