Research On Intrusion Detection Of High-speed Railway Perimeter Based On Optical Fiber Sensing And Video Cooperation

With the rapid development of China’s railway and intelligent transportation system,the transportation mileage of high-speed railways continues to increase,creating convenience for people to travel.However,due to the long line along the railway and the complex environment around the railway,some dangerous intrusion behaviors inevitably occur in the high-speed railway perimeter.How to monitor the security of high-speed railway trains and accurately detect perimeter intrusion behaviors has become an important issue to be solved today.Compared with the existing high-speed railway perimeter intrusion detection methods,fiber optic sensing monitoring technology and video camera monitoring technology are easy to implement,low in cost,wide in detection range,and can achieve precise positioning,which is more suitable for abnormal intrusion detection in the high-speed railway perimeter.In this thesis,an optical fiber sensing and video image collaborative intrusion detection method is proposed with high-speed railway perimeter intrusion detection as the research background.This method effectively solves problems that current intrusion detection methods do not fully extract the Spatio-temporal features of intrusion behaviors and have low robustness to light changes and environmental background interference.Evaluation on test experiment proves that the method can detect four kinds of high-speed railway perimeter intrusion behaviors and the detectionF₁-score of each intrusion behavior is above 95%,and its performance is higher than the existing intrusion detection methods.In addition,the method can distinguish the non-threatening background noise interference generated by the passing high-speed railway trains,which better meets the practical needs.The main work accomplished in this thesis is as follows:（1）The intrusion behavior detection algorithm based on optical fiber sensing is deeply studied,and a distributed optical fiber intrusion behavior detection system based on Faster R-CNN is designed.Firstly,the fiber optic signals around the high-speed railway are collected using the Distributed Acoustic Sensing（DAS）system based on Phase-sensitive Optical Time Domain Reflectometry（Φ-OTDR）.Then the collected fiber optic signals are analyzed and processed in time and spatial dimensions to form intensity images with Spatio-temporal signal features.Finally,the Feature Pyramid Network（FPN）-based Faster R-CNN algorithm is proposed,which enables the algorithm to extract the features of multi-scale fiber optic signal intensity images to improve the detection accuracy of the system and meet the demand of high-speed railway perimeter intrusion detection.（2）Aiming at the defects of the current intrusion detection methods based on video images,the human skeleton sequence in the video images is modeled by using the features of the human pose skeletal joint points and a video image intrusion behavior detection algorithm based on Spatial Temporal Graph Convolutional Network（ST-GCN）in high-speed railway perimeter is proposed.The Spatio-temporal relationship between human pose behaviors is used to detect high-speed railway perimeter intrusion behaviors in video images to enhance the algorithm’s robustness.At the same time,a multi-sensor data association algorithm is designed,which uses the time interval information of the intrusion behavior and the time data of the intrusion behavior detected by the optical fiber sensing and video sensing,to correlate video sensing with fiber optic sensing covered by its shooting distance.This correlation makes the two kinds of sensor data complement each other to jointly detect the intrusion behaviors in the high-speed railway perimeter and obtain more accurate and stable detection results.