Research On Railway Foreign Object Intrusion Detection Algorithm Based On Computer Vision

With the continuous growth of China’s high-speed railway operation mileage,the operation safety of high-speed railway has attracted much attention.Due to the high speed of high-speed railway,once personnel or other foreign matters invade the railway line,it will pose a great threat to the operation safety of railway trains.Therefore,real-time monitoring of the railway and judging whether there is railway foreign matter invasion through the target detection algorithm can eliminate potential safety hazards,reduce the incidence of dangerous accidents and ensure the operation safety of trains.In recent years,with the increasing number of monitoring facilities along the railway and the rapid development of video monitoring technology,the combination of computer vision technology and monitoring images under the railway scene has realized the foreign object detection and real-time monitoring of the railway operation environment.This method can not only improve the work efficiency,save labor cost,but also ensure the all-weather monitoring of railway operation.Therefore,using the method based on computer vision to detect the foreign matter invading the railway is of great significance to ensure the safety of railway operation.Based on the analysis of the research status of target detection algorithm and railway foreign object detection at home and abroad,aiming at the existing problems of railway foreign object intrusion detection,this paper makes a relevant in-depth research on Railway foreign object intrusion detection algorithm.The main contents include three aspects: Research on lightweight railway foreign object intrusion detection algorithm based on interest region division Research on foreign object detection algorithm of railway with dense attention in infrared weak light environment and foreign object detection algorithm of railway occlusion based on multi-scale context information.The main research contents are as follows:(1)Aiming at the problems of error warning,low detection efficiency and large amount of model parameters in the current railway foreign object intrusion algorithm based on computer vision,a foreign object intrusion detection method in railway area of interest based on lightweight network is proposed.Firstly,the method of perspective transformation and cubic function fitting is used to detect the railway line,and the region of interest of railway foreign object intrusion detection is obtained.Then,the thinning and channel pruning methods are used to compress the YOLOv3 model,and a lightweight railway foreign object detection model is constructed.Finally,the railway data set and field experiments show that the parameter space of the lightweight model of the proposed algorithm is effectively reduced,and the detection speed is 1.6 times that of the original YOLOv3 model.It can quickly and effectively detect the foreign object intrusion in dangerous areas in different railway scenes and reduce the error warning.(2)Aiming at the problems of insufficient target feature extraction and low detection accuracy in the detection of railway foreign matter intrusion in the infrared weak light environment,a dense attention railway foreign matter detection method in the infrared weak light environment is proposed based on the Mask R-CNN detection model.Firstly,a densely connected multi-scale FPN pyramid network is proposed to strengthen the use of feature map,so as to improve the detection accuracy in infrared weak light environment.At the same time,CBAM attention mechanism is introduced and Res Net-FPN network structure is improved to improve the attention to the target area and highlight the target characteristics in the infrared weak light environment.Secondly,the k-means algorithm is improved to re preset the size of the anchor box to improve the accuracy of the anchor box in locating the target area.Finally,it is tested by railway infrared data set and field experiment.The experimental results show that the proposed method has high detection accuracy,with an accuracy of 89.24%,which is 6%higher than that of Mask R-CNN,and the pixel accuracy is 8%.This method can more accurately detect railway foreign matters in the infrared weak light environment,and can realize the division of railway clearance area.It is better than the comparison method in subjective and objective evaluation.(3)The target sizes of foreign objects invading railway gauge are different,and often accompanied by the phenomenon that multiple targets block each other,resulting in the incomplete characteristic information of foreign objects,resulting in the problem of missing detection of railway foreign objects.To solve this problem,a railway occlusion foreign object detection algorithm based on multi-scale context information is proposed.Firstly,Dense Net dense connection network is used as the feature extraction network to extract feature information more effectively.At the same time,the context information module is integrated into the backbone network to supplement the occluded target information with the help of global context information;Then,according to the changeable shape and scale of railway foreign object target,a multi-scale feature fusion network suitable for railway foreign object detection is designed to make use of the effective information of different receptive fields;Finally,the prediction network of FCOS algorithm without anchor frame is used to realize the classification and location regression of targets.The experimental results show that the network model detection of the proposed algorithm is accurate and fast,and good detection results are obtained on the self built railway foreign object data set.The average accuracy reaches 78.46%,and the missed detection rate is lower than that of the comparison algorithm,which has a good effect of railway occlusion foreign object detection.