Algorithm Research And System Design Of High Performance Pedestrian Detection For Railway Security

Railway transportation is the main artery of our national economy.Traffic safety is the lifeline of railway transportation.It not only affects the production efficiency and economic benefits of enterprises themselves,but also has a significant impact on society and economy.Invasion of pedestrians in key railway areas is an important factor affecting traffic safety.At present,artificial patrol or simple alarm devices are still widely used to realize intrusion detection on railway platforms and along railway lines,but these measures have cost and time limitations.Therefore,the railway system needs intelligent equipment to realize pedestrian detection and recognition,in which pedestrian detection is the fundamental and important part.Based on the analysis of the application of railway scenes,this paper proposes a pedestrian detection algorithm based on deep learning.The main contributions of this paper are as follows:(1)To solve the problem of small-scale pedestrian detection in railway scenes,a small-scale pedestrian detection algorithm based on Faster R-CNN network is proposed.The network is optimized from three aspects:weak semantic segmentation loss function,channel affine transformation and feature fusion.Test results on public datasets and a private dataset of railway application scenarios show that the proposed algorithm can significantly reduce the miss rate of small-scale pedestrian detection.(2)In order to further reduce the false positive rate of small-scale pedestrian detection,a detection framework that utilizing motion information to filter bounding boxes is proposed.The motion information of pedestrian that estimated by the structural similarity between video frames,combined with the confidence score of the network detection result,can be used to reevaluate bounding boxes.Meanwhile,bounding boxes are propagated and restrained among frames while the small-scale pedestrian detection results are effectively reserved.Test results show that the framework can apparently reduce the false positive rate and improve the overall operational efficiency.On the hardware platform of video acquisition and processing with a zoom camera,a pedestrian detection system for railway scenes is designed and implemented using PyQt5.Video acquisition and processing test on NVIDIA Jetson TX2 proves that the system can realize detection and tracking function well and can be used in pedestrian intrusion detection in railway scenes.