Robust Multiple Target Tracking

ObjectVideo and University of Southern California propose to develop innovative algorithms for robust and efficient visual tracking. The key issues of multiple target tracking are: variations in target shape and appearance due to camera viewpoint and illumination condition, non-linear target motion, self and inter-occlusion, image noise and distortion, and high scene clutter. We identify four main goals to ensure robust tracking: (i) accurate detection of targets, (ii) robust online tracking, (iii) persistent tracking of multiple targets across multiple frames, and (iv) tracking across multiple cameras. In Phase I, we developed algorithms for target detection based on the AdaBoost technique using different types of image features. We designed a collaborative multiple-kernel algorithm for robust online tracking as well as a multi-frame algorithm for data association. In Phase II, we will extend our approach to handle more complex and challenging environments, such as low frame-rate and noisy video. We will also develop algorithms to perform cross-camera tracking and to process high-resolution video. The product of this effort is a technology demonstration system that performs robust real-time multiple target tracking in challenging environments.