Research On Obstacle Detection For Railway Tracks

Railway tracks obstacle detection is an important measure to ensure the safety of railway tracks.In the existing machine vision-based railway obstacle detection method,a rail transit obstacle detection method that aligns a reference image without an obstacle and a target image that may have an obstacle and compares the difference between the images is novel and effective.Due to the effects of train speed changes,jitter and illumination changes,this method is susceptible to alignment differences,illumination changes.Studying more effective and accurate methods for detecting railway obstacles has urgent practical needs.Based on the comparison of current domestic and foreign railway track obstacle detection research,the thesis proposes a railway track obstacle detection method.Firstly,Detecting track boundary areas based on track characteristics.Sencondly,real-time matching of the target image captured during the train driving based on the pre-acquired reference video sequence without obstacles.Lastly,illumination robust obstacle detection method to achieve fast and accurate detection of obstacles.The main contents of this thesis include:(1)For the problem of orbital bounding area extraction,the method of track segmentation and path growth is adopted.Detection of near-field straight track by using Hough transform based on finite prior knowledge,the near-field straight track will determine the start point of the far-field track.In order to detect the far-filed track,the method of segmentation detection of the far-filed track will be adopted.Firstly,the path based on the angle is used to score the path of growth,and the optimal growth path is selected based on the score.Based on the track detected in the image and the specified limit range equally extracted track bounds.The method can realize track detection and track bound region extraction under illumination variation and curvature variation.(2)This thesis improves the time-space alignment problem between the reference video sequence and the target video sequence caused by train speed variation and jitter.A continuous video image distance metric based on normalized downsampling is used,and the online dynamic time warping algorithm is improved for time alignment.The inner point set extraction area of the random sampling consistency algorithm is defined for spatial alignment,and the reference image can be realized.Matches the target image in real time,and the orbital bounds of the two images coincide.Realize that the reference image matches the target image in real time.(3)In order to solve the influence of illumination on obstacle detection in the orbital bound area,an obstacle detection algorithm for aligning orbital traffic images with illumination robustness is adopted.By effectively combining the image block difference information,the image local feature difference information and the image color space distance information with illumination robustness,the possible existence area of the obstacle is detected.Based on the principle that orthogonal decomposition of tiny image blocks effectively reflects illumination changes,the maximum singular value is normalized to eliminate the effects of illumination.Finally,based on the possible areas of the obstacle,the obstacle detection of the illumination is robust and fast.All the above work uses real railway track scene data for experiment and analysis.The algorithm in this thesis can be used to effectively and accurately detect railway obstacles.