Design Of Railway Crossing Video Surveillance System And Research On Multi-target Tracking Algorithm

With the continuous intensification of China’s railway network,railway traffic accidents have become increasingly apparent.Railroad crossings are the only conflicting areas of right-of-way in the entire railway transportation network,and they are more prone to safety accidents.Based on the intelligent reconstruction project of a seaport railway crossing,this paper studies and designs a port railway crossing intelligent monitoring system,and conducts research on multi-target tracking algorithms in combination with monitoring scenarios.After conducting on-site survey and monitoring needs analysis of crossings,three video surveillance systems for complex crossings were designed,and a video monitoring platform for the monitoring center was constructed.Based on the research and improvement of Faster R-CNN deep learning network,the crossing detection and detection of vehicles were realized.Recognition;Research on SORT algorithm to achieve multi-target tracking of crossing vehicles;and use graphics to formulate vehicle trip line rules to analyze the behavioral alarm of tracking vehicles.Automatically alarms when the target is still tripped when the crossing is closed,improving the intelligence of video surveillance.The main work of the paper is as follows:(1)Design of Railway Surveillance Video Surveillance System:This paper analyzes from the two aspects of the monitoring system’s organizational structure and functional modules,and uses the server and the operating display as the two design ends to design a set of railway crossing intelligent video monitoring systems that meet the requirements of railway specifications.(2)Improved Faster R-CNN network target detection and recognition:This paper conducts algorithm research on crossing vehicles,and proposes a Density Faster R-CNN network model combining density statistics to improve the target detection rate under vehicle occlusion.After several comparative experiments,the effectiveness of the improved algorithm proposed in this paper for vehicle detection at crossings was verified.The overall m AP of the detection model reached 81.45%,and the detection speed was 8fps.(3)Target tracking and cross-line alarming based on SORT tracking algorithm:This paper uses the Density Faster R-CNN detection algorithm and the SORT tracking algorithm to perform multi-target tracking tasks on railway crossings.The experimental results show that multiple tracking evaluation indicators can increase when the occlusion recognition rate is improved.Two trip-trip incidents were formulated to implement a complete "detection + tracking + behavior determination" process for cars,SUVs,trucks,vans and motorcycles at railway crossings.(4)Data set for railway crossing target detection and tracking:The data set contains two parts: the target detection picture data set and the tracking video frame data set.Relying on the project during the installation and commissioning and trial operation phases,three road crossing videos were collected and vehicle data sets were self-made,including 6,526 picture sets and 13 sets of videos totaling 11 G.(5)This article relies on the intelligent transformation of a seaport railway crossing to install and update the monitoring equipment and design the system monitoring interface for the three crossings.Experiment The intelligent monitoring system and detection algorithm designed in this paper are used to determine and record the traffic behavior.The system can work normally for half a year after trial operation.