On collaborative people detection and tracking in complex scenarios

摘要

The main contributions of this paper covers two different aspects: people detection and tracking. A whole detection/tracking system that integrates appearance, motion and tracking information is presented. This system uses the information provided by each of the independent tasks to improve the final result of the system. The tracking information is integrated in the detection task improving the detection results and vice versa. The experimental results over an extensive and challenging video dataset point out the state of the art limitations in complex or realistic scenarios, and show that the proposed collaborative system significantly reduces these limitations and improves the results in this kind of scenarios.