Research On Algorithm Of Detection And Recognition For Obstacle In Front Of The Train Based On Monocular Vision

The significant mission of national mobility and freight transportation is loaded by railway.China's railway not only has a long mileage,but also crosses different terrains.Frequent invasion events of railway clearance are caused by natural disasters such as landslide,collapse,debris flow,and random obstacles for instance the front trains,pedestrians,object,etc.In the high-speed driving environment,it is difficult to implement the real-time and reliable monitoring of the driving environment based on the driver's visual,trackwalker and single point monitoring.With the rail transit automatic driving signal system research and development and the application of railway visual monitoring technology,intelligent traffic monitoring in front of railway clearance environment has become a hot research topic,and possesses important practical significance.The detection and recognition of obstacles in front of train is used on-board monocular vision method to achieve,based on the current domestic and foreign existing rail clearance monitoring technology research and comparison.According to the analysis of the target sequence frames,relying on track environmental characteristics,three key steps are focused on solving,namely the identification and tracing of the railway clearance area,location and extraction of suspected foreground obstacles and the detection and recognition of invasion obstacles.The main contents of this thesis include:(1)Study on detection and tracking algorithm of railway clearance.Aiming at the problem of insufficient accuracy in the current research on static linear railway clearance model,the piecewise switchable curve model is proposed to track and describe the track.The railway line is taken as the reference,and the pixel scale transformation is used to gauge calibration range.The improved Hough transform is adopted to solve the parameters of straight rail in close area.Based on the Hu invariant moments,the search algorithm of the moving window is proposed to get the feature points of railway tracks,and the model is fitted by these points.Then,according to the real-time update the model switching strategy and affine transform principle and scope.Railway clearance range is updated in real time through the model switching strategy and principle of affine transformation.(2)Research on Algorithm of target localization for suspected obstacles.At present,the foreground object detection method has large interference from background in the dynamic environment of the track.In this thesis,based on the frame difference method,according to the characteristics of the inter track texture,the mathematical morphology is introduced to reconstruct the inter track texture and to realize the difference compensation.The location and segmentation extraction of the target is determined by the saltatorial value of gray difference accumulation value in the target position of the suspected obstacle.(3)Research on obstacle detection and recognition algorithm.In view of the present study,existing problems include that characteristics of target identification criterion is simple and the changing environment of the track is difficult to adapt.In this article,the main objects of interference obstacle target identification are analyzed,and many kinds of shape features and texture features are targeted to enrich the feature types.Adaboost algorithm is adopted to realize the identification of suspected targets,and it learns to get a strong classifier which is integrated by the classifier of multi class features.The accuracy of recognition is improved effectively.The experimental analysis in the different scenarios shows that this algorithm can effectively improve the detection accuracy in real-time processing,and has a good ability to adapt to different environments.