Person Re-Identification by Deep MAX Pooling Network

Person re-identification is a fast growing research area that aims at matching pedestrian images across camera views. This problem is particularly challenging because of complex variations of viewpoints, poses, lighting, occlusions, resolutions, background clutter and camera settings. A Person re-identification algorithm based on the deep max pooling network is proposed for the difficult problem. First, we present a novel Convolutional Neural Network (CNN) called deep max pooling network to learn features of the input persons, and then the algorithm gets a similarity measure function of the related person with independent metric learning. Finally, the algorithm weights the original similarity and gets the ultimate similarity. The algorithm proposed in this paper can effectively express pedestrian image information. Furthermore, the proposed method has strong robustness to variations of viewpoints, poses, lighting, occlusions, resolutions, background clutter and camera settings. The proposed method achieves a 69.5% rank-1 on market1505 benchmark. It greatly improves the recognition rate and has practical application value.