Double-Line Multi-scale Fusion Pedestrian Saliency Detection

Abstract

Pedestrian salient detection aims at identifying person body parts in occluded person images, which is greatly significant in occluded person re-identification. To achieve pedestrian salient detection, we propose a double-line multi-scale fusion (DMF) network, which not only extracts double-line features and retains both high-level and low-level semantic information but also fuses high-level information and low-level information for better complement. CRF is then used to further improve its performance. Finally, our method is used to deal with occluded person images into partial person images to achieve partial person re-identification matching. Experiment results on five benchmark datasets show the superiority of our proposed method, and result on two occluded person re-identification datasets indicate the effectiveness of our proposal on pedestrian salient detection.