A Shunting Safety Risk Identification Method Based On Fusion Of Radar And Vision

Shunting operation is an important component of railway freight transportation and an essential part of ensuring efficient and stable railway production.Shunting operations have the characteristics of complex environment,busy lines,and intensive operations.The composition of the train during shunting operations changes in real-time to meet transportation needs.Currently,shunting operations rely on manual operation,and due to the relatively open working environment,there are many safety risks and difficulties in identifying them.Drivers and shunters need to maintain a high level of attention,identify and respond to potential hazards.The difficulty in identifying safety risks makes it hard to effectively recognize collision,derailment,and other accidents,which greatly affects transportation safety and operational efficiency.Therefore,it is of great significance to conduct research on the method of identifying safety risks in shunting operations.This dissertation focuses on safety risks identification during the shunting operation process and proposes a radar and vision fusion method to quickly identify and classify obstacles in front of the locomotive.This method addresses the limitations of single sensor application conditions,insufficient reliability,and low accuracy.The main contents of the thesis are as follows:(1)Feature-based shunting safety boundary construction method.To reduce resource expenditure and effectively identify potential safety risks in the shunting process,combining the characteristics of multiple shunting routes with significant differences,this dissertation proposes a Canny algorithm based on dynamic threshold adjustment and a steel rail detection algorithm using probabilistic Hough transform to address the problems of poor anti-interference ability and insufficient real-time performance in current visual rail detection algorithms.By defining the safe area for train operation,this method has been experimentally validated for its effectiveness and accuracy.(2)A target detection method based on improved YOLOv5 s.Given the complexity of the environment and the timely identification of security risks during the shunting operation,an improved Mosaic data augmentation method is proposed for YOLOv5 s to enhance the model’s generalization ability.The detection capability for small targets is improved by adding a target detection layer.Attention mechanism is introduced to strengthen feature fusion,improving detection speed and accuracy.The effectiveness of the improved YOLOv5 s is verified by experiments.(3)Data layer fusion-based multi-sensor fusion method.To solve the reliability problem of single data,an improved multi-sensor fusion algorithm based on data layer fusion is proposed.Considering the errors and redundancies in the data collected by sensors,a method of matching local nearest neighbor data frames is proposed to increase the credibility of information and obtain a consistent description of the target to be tested.This method achieves time synchronization of multi-source data.By conducting experiments,the coordinate transformation matrix between the radar coordinate system and the image coordinate system is obtained,and unified coordinate representation achieves spatial synchronization.The boundary threshold for safe train operation is introduced,and a radar data filtering algorithm for the target’s dynamic survival cycle is proposed,which reduces radar interference information and multisensor data fusion errors.The effectiveness of the fusion algorithm was verified through data fusion testing experiments.There are 57 pictures,13 tables and 100 references.