Person Re-identification Based on Multi-level Feature Fusion with Overlapping Stripes

Abstract

Most of the local features based person re-identification (Person-ReID) methods have the problem of lack of robustness and discriminability due to the distortion of information in the procedure of extraction of pedestrian features. In this paper, a novel Person-ReID algorithm based on multi-level feature fusion with overlapping stripes is proposed. In the training process, the output feature maps from different layers of the backbone network are segmented equally in the vertical axis. The features with overlapping stripes are extracted to compensate the loss of information. Three different loss functions are used for different feature vectors in the procedure of the training to minimize the intra-calss distance. Group normalization modules are applied to reducing the optimization differences within various loss functions for obtaining appropriate shared features. In the inference stage, multiple feature vectors are fused into a new feature vector, and the similarity is calculated. This algorithm is performed on Market-1501 and DukeMTMC-reID datasets with the analysis of experimental results. The proposed algorithm can improve the accuracy of Person-ReID, and the features extracted by the model have strong robustness and discriminability.

Keywords

• feature with overlapping stripes
• feature fusion
• joint learning
• person re-identification (person-reid)
• deep learning