Graph Neural Networks for Cross-Camera Data Association

Cross-camera image data association is essential for many multi-camera computer vision tasks, such as pedestrian detection, multi-target tracking, 3D pose estimation, etc. This task typically modeled a bipartite graph matching problem and often solved by applying minimum-cost flow techniques, which may be computationally demanding large data. Furthermore, cameras are usually treated pairs, obtaining local solutions, rather than finding global solution at once all multiple cameras. Other key issue that of the affinity function: widespread usage non-learnable pre-defined distances, Euclidean Cosine ones. paper proposes an effective approach cross-camera data-association focused on solution, instead processing pairs. To avoid fixed distances thresholds, we leverage connectivity Graph Neural Networks, previously unused in this scope, using Message Passing Network to jointly learn features similarity functions. We validate proposal association, showing results over EPFL dataset. Our considerably outperforms literature without requiring trained same scenario it tested. code available https://www-vpu.eps.uam.es/publications/gnn