Multi-Camera Trajectory Based Vehicle Re-Identification for Robust Traffic Monitoring

Vehicle Re-Identification (ReID) is the challenging task of identifying the same vehicle across multiple cameras in urban environments. In this paper, we propose a novel MultiCam TrajectoryID model that leverages trajectory-based features, addressing the limitations of single-image-based ReID methods. The system combines feature extraction methods like SIFT with the Kalman filter and You Only Look Once (YOLO) algorithms for trajectory prediction to guarantee accurate and effective tracking of objects across several camera views. This MultiCam TrajectoryID approach enhances traditional single-image vehicle ReID by focusing on the continuity of object movement, improving the robustness of identification in complex, real-world traffic environments. We evaluate the performance of our model on the challenging VeRi-776 dataset and compare it with state-of-the-art methods. The proposed model achieves a mean average Precision (maP) of 86% on VeRi-776 data, outperforming recent approaches such as Multi-Branch Enhanced Discriminative Network and Multi-Fine Grained Network models. This significant improvement demonstrates the effectiveness of trajectory-based re-identification in multi-camera setups and its contribution to vehicle ReID and real-time surveillance systems.