Distance Aggregation based Score Fusion for improving person re-identification

Person re-identification is an important problem for automated video-based surveillance. Person re-identification is the problem of identifying a person over multiple cameras with varying viewpoints at different locations and different points of time. The most common directions of research in this domain are feature extraction and distance metric learning. While a number of algorithms have been proposed for solving this problem, the performance of these algorithms is varied and mistakes in detecting the correct person are often encountered. In this paper, we propose a novel framework for aggregating the “distances” between images of different persons that are generated by person re-identification algorithms in order to rank “gallery” images in terms of their “closeness” from a given “probe” image. The Distance Aggregation-based Score Fusion (DASF) framework uses aggregated distances to rank the gallery, and results in improved performance over each individual person re-identification algorithm that is used to obtain distances. Towards this end, we have applied Simulated Annealing, a heuristic algorithm, to find the optimum distances between images using the distances from individual algorithms. The DASF framework is unsupervised and requires no prior information about how each of the individual algorithms works. Our results show significant improvement in performance for commonly used datasets. Performed experiments also indicate that the DASF framework yields better results than the state-of-the-art score fusion algorithms for person reidentification.