Pedestrian detection based on improved Faster RCNN algorithm
Xiaoqian Yu, Yujuan Si, Liangliang Li · 2019
Pedestrian detection is receiving more attention with the development of deep learning and smart driving technology. However, the performance of existed pedestrian detection schemes are influenced by complicated natural scenes. To address the problem, we improve the Faster RCNN original framework by combining feature concatenation and hard negative mining strategies to boost the performance on challenging pedestrian detection datasets. In this paper, the proposed model is trained on Daimler pedestrian dataset and then tested on public Caltech and INRIA pedestrian datasets, which achieves the miss rate of 24.27% and 10.31%. The results indicate that the improved Faster RCNN network outperforms the state-of-the-art methods.