Robust Patch-Based Pedestrian Tracking Using Monocular Calibrated Cameras

Although several methods for pedestrian tracking can be found in the literature, robustly tracking a person in unconstrained environments is an open and active research problem. In this paper, we propose a method that represents each pedestrian as a set of multiple fragments, aiming robustness with respect to occlusions. These patches are tracked individually and their translation vectors are combined robustly in the world coordinate frame using Weighted Vector Median Filters (WVMF). Additionally, the algorithm uses the camera parameters to both estimate the person scale in a straightforward manner and to limit the search region used to track each fragment. Experiments carried out using two publicly available datasets (PETS and Town Centre) are presented, and they indicate that the proposed method is robust to partial occlusions and large scale changes. According to our experiments, the proposed approach outperforms, regarding the quality of localization, some of the methods in the current state of the art.