Person re-identification under the problem of path selection

In this paper, a novel person re-identification method is introduced under the problem of path selection. Unlike other supervised person re-identification algorithms, our method can identify persons in the video without artificial markers. Our method divides each of the image sequences into several slices and selects the most distinguished ones, which can improve the performance. More crucially, this model is unsupervised and readily scalable to real-world large scale ReID settings, and more suitable to previous path selection, i.e., no need of exhaustively collecting large numbers of cross-view pairwise labels for each camera pair, as required by most existing ReID models, which need supervised training. Experimental results show that our method can outperform other methods and achieve excellent results.