부산대학교 도서관

Face recognition and tracking technology have witnessed significant advancements during the recent decades. These advancements include improved accuracy and speed in identifying and tracking individuals, as well as the ability to recognize faces in various scale and lighting conditions. Leveraging the potential of face recognition and tracking, this article explores the integration of process mining techniques to discover and visualize face’s appearing trajectories in crowd scenarios, aiming to enhance crowd security, surveillance, and personal identification. Notably, existing face recognition tools typically focus on bounding box localization, neglecting the utilization of face coordinates to construct trajectory models upon face re-identification. In this paper, full system architecture for building a trajectory model of re-identified faces in a crowd is proposed. This approach significantly helped in building a large database of visitor faces, and the proposed trajectory model resulted in a high rate of true positive face re-identification.