An Improved Faster R-CNN Algorithm for Pedestrian Detection

Abstract

Pedestrian detection is an important branch of computer vision and has been the focus of research due to its wide range of applications. Although commonly used object detection model Faster R-CNN has achieved good results. However, there are still some shortcomings in the specific task of detecting pedestrians. This paper made three improvements to the Faster R-CNN to better adapt it to the pedestrian detection task. First, we did a lot of experiments and finally chose MobileNetv2 as our backbone network. Second, we designed a multi-branch feature pyramid network (M-FPN), which is used to better integrate the model's shallow feature information with the deep feature information improved the model's ability to detect pedestrians. Finally, an attention region proposal network SE-RPN is used to improve the model's ability to focus on pedestrian features and suppress attention to background interference features. The experimental results show that the improvement strategy proposed in this paper has achieved better results. These strategies improve the average accuracy of Faster R-CNN on our self-built dataset by 6.14% and the detection speed by 27fps. The AP on Caltech dataset reaches 87.01%, and the detection speed can achieve 39.4fps.