Research On Detecting Intrusion Limits Of Foreign Objects In Railway Based On 2D Laser Radar And Vision

In recent years,the development of China's railway by leaps and bounds has raised new challenges for train safety driving protection systems.At present,the detection system of foreign body invasion and limitation of railways mainly includes contact and non-contact methods,including dual-grid,optical fiber sensing,and computer vision.The computer vision method has the characteristics of large field of view and easy installation.However,because the image quality is easily affected by light and the environment,the system has poor robustness and low reliability.The laser radar has strong anti-interference ability,does not depend on the external light environment and has a short reaction time,and can compensate for the lack of the camera.However,the data information obtained by the two-dimensional laser radar scanning obstacles is single,and is not as rich as the target detection information of the visual detection.Therefore,this paper proposes a system for intrusion detection of foreign matter(mainly pedestrians and rockfalls)that incorporates 2D laser radar and visual inspection.Since the visual detection algorithm has been relatively mature,this paper focuses on the integration of target detection,tracking,and visual inspection results based on two-dimensional laser radar.The specific research content is as follows:(1)In-depth analysis of the principle and characteristics of two-dimensional laser radar,and visualization of the laser radar data.Based on the characteristics of two-dimensional laser radar data,that is,monitoring environmental background points in a straight line,this paper uses Hough transform to detect and extract the background so that the background is removed in each subsequent frame of data.(2)Aggregation Hierarchical Clustering(AHC)algorithm is used to cluster front site and detect obstacle targets.Since the lidar data is discrete,and there are noise points in the previous scenes in addition to the target data,further analysis and judgment are required after clustering.This paper proposes a method for distinguishing the subject from noise based on the number of relative points.(3)After the target detection is completed,the Kalman tracking algorithm is used in the target tracking module.This algorithm can solve the problem of missing due to target occlusion in the multi-target tracking scenario,and the tracking effect is obvious.(4)Calibrating two different sensor devices,two-dimensional laser radar and camera,and adopting Perspective Transformation(PT)algorithm to determine the conversion relationship between the two coordinates.Simulate the monitoring scene in the lab hall,fuse the target detection results of the two devices,and use the Distance Transformation(DT)algorithm to highlight the fused target.The experimental results show that the fusion of the data collected by the two sensors can more accurately realize the detection of intruded foreign objects.