Re-Identification for Multi-Target-Tracking Systems Using Multi-Camera, Homography Transformations and Trajectory Matching
Dominik Heimsch, Yan Han Lau, Chinmaya Mishra, Sutthiphong Srigrarom, Florian Holzapfel · 2022
This paper proposes an algorithm for multiple-target re-identification of targets viewed using images captured from multiple cameras. B-spline polynomials are used to approximate and define the target trajectories. This allows a smoother representation of irregular trajectories of rapidly moving targets. Homography transformations of the different camera frames into one common frame allow a comparison between target trajectories. Based on their similarity targets are re-identified, meaning that corresponding targets observed by multiple cameras are associated across cameras. Results show that the algorithm is able to re-identify targets in a simulated environment as well as on recorded data of football players from field tests.