Person Re-Identification Based On Convolutional Neural Network

Person re-identification,i.e.given a pedestrian image from a camera,recognizing the same person from candidate pedestrian images shot by other spatially disjoint cameras.With the rapid growth of video surveillance,the practical application value and theoretical research significance of person re-identification in the field of intelligent video surveillance is increasing remarkably.Various traditional methods have been proposed to solve person re-identification problem.Among these studies,researchers mainly focus on hand-crafted pedestrian feature extraction and distance metric learning.In this paper,we propose a novel algorithm based on convolutional neural network to solve person re-identification problem.In the convolutional neural network,we combine pedestrian deep feature learning and distance metric learning to achieve state-of-art person re-identification accuracy.In practical terms,we design a convolutional neural network spe-cialized for pedestrian deep feature learning,and we apply clustering loss function proposed by us in the loss layer.Clustering loss function can reduce intra-class distance and increase inter-class distance.Therefore clustering loss function can improve the discrimination of deep feature.Experiments on six common person re-identification data sets demonstrate the great superiority of the proposed algorithm.