Multi-scale Pedestrian Detection Based on Attention Mechanism and Feature Fusion

Abstract

Scale variation of pedestrian targets is a major challenge in pedestrian detection, which leads to difficulties for pedestrian detection algorithms to accurately capture pedestrian targets at different scales. To address the above problems, this paper proposes a multi-scale pedestrian detection method based on attention mechanism and feature fusion. First, a new feature fusion module is constructed to improve the problem of insufficient semantic information of shallow features, so that the feature information of different scales can be fully fused to strengthen the detector's feature extraction ability for small-scale target pedestrians. Second, we introduce a spatial channel attention mechanism in the network to suppress irrelevant background information and enhance the extraction of key feature information of pedestrian targets. Finally, we optimize the original prior box parameters to generate more suitable prior boxes for detecting pedestrians to improve detection accuracy. Comparison experiment results on Caltech-USA and CityPersons pedestrian detection datasets show that our method achieves very competitive performance with the state-of-the-art methods.