"Person Re-identification Based CNN and Score Normalization"

Abstract

Person re-identification (Re-ID) is an important task in the field of intelligent security. It’s defined as the problem of identifying people across images that have been captured by different surveillance cameras without overlapping fields of view. In a real world environment, Re-ID is a challenging task due to variations in lighting conditions, viewing angles, pose and occlusions. Two fundamental problems are critical for person re-identification, feature representation and metric learning. An effective feature representation should be robust to illumination and viewpoint changes, and a discriminant metric should be learned to match various person images. In this work we are going to use an effective feature representation called convolutional neural network (CNN) to detect patterns to distinct persons, and metric learning method called crossview quadratic discriminant analysis (XQDA) with the mahalanobis distance to improve classification results. Then we will apply a score normalization on the resulting mahalanobis distance on a four challenging person re-identification datasets VIPeR, GRID, CUHK01 and PRID450S. Experimental results on the four mentioned datasets demonstrate the effectiveness of the proposed feature.