Multi-scale Pedestrian Detection Based on Receptive Field Matching

Abstract

Pedestrian detection is a hot topic in both academia and industry. However, the pedestrian in the image presents different scales due to the different distance from the fixed cameras. Thus, how to detect pedestrians of different scales has become an attractive problem in pedestrian detection field. In this paper, we propose a simple and compact multi-scale pedestrian detection architecture based on receptive field matching (denoted as RFMNet), which can cover continuous multi-scale pedestrians with 100% in theory. In this model, the receptive field is regarded as an invisible “anchor”, so that the feature points with different scale receptive field can detect different scale pedestrian targets without any bells and whistles. Based on the above analysis, the proposed approach has an anchor-free setting. The extensive experiments on Caltech-USA benchmark demonstrate that our method outperforms the state-of-the-art pedestrian detection algorithms.