Research On Algorithm Of Intrusion Detection For Railway Orbital Foreign Object Based On Machine Vision

Railway transportation is an important infrastructure for public travel and cargo transportation.With the rapid development of China's railway construction,the increasing mileage and the continuous improvement of the road network,the railway covers a wider range of terrain,which makes the environment of train running more complex.The frequent limit intrusion accidents seriously threaten the safety of train running.It is of great significance to realize the detection of railway intrusions and effectively avoid the accidents of invading limit of intrusions.The existing safety monitoring measures can not meet the traffic safety protection for the violations caused by pedestrians,transportation vehicles,construction waste,and natural disasters,such as rock fall and debris flow.Studying the existing rail transit clearance monitoring technology at home and abroad,machine vision combines with image processing technology has become a hot research topic in traffic safety monitoring.In this thesis,a machine vision based intrusion detection algorithm for orbital foreign object is proposed.The research content mainly includes:(1)Division of railway transgression area.In view of the influence of ballast,sleepers and other noises,the traditional edge detection algorithms have poor effect on rail-edges detection,and the invasion zone on the basis of rail-edges cannot be accurately divided.In this thesis,a rail-edges detection algorithm based on gray-scale morphology is proposed to realize the adaptive detection of rail-edges,and the result is optimized by using multi-constraint Hough transform,the detection result is optimized by using the multi-constraint Hough transform to realize the accurate division of the invasion area.(2)Detection of intrusions and judgment of intrusions behavior.First of all,this thesis studies the commonly used intrusion detection algorithms of railway,aiming at the problems of missing detection and wrong detection in the traditional target detection algorithms,this thesis improves the Vibe(Visual background extractor)target detection algorithm to accelerate the elimination of ghost in the initialization of background model,and realizes the detection of intrusion target according to the adaptive segmentation threshold.In order to complete judgment of the ghosts,the improved three frame differential accumulation ghost detection algorithm is proposed.Then,the improved Vibe algorithm is used to initialize the KCF tracking algorithm.By using KCF(Kernelized Correlation Filters)tracking algorithm to achieve the continuous tracking of the intrusion target and determine the distance from the intruder to the center line of the orbit to judge whether the intrusion occurs.(3)Identification and classification of intrusions.Aiming at the problem that the distinguishing features of intrusions are single in the current study of intrusion recognition and classification and difficult to adapt to the changeable environment of railway,firstly,the haar-like feature and texture feature of the candidate region of the intrusion are extracted,and then by using Adaboost algorithm to select the features that can accurately describe the types of intrusions to form strong classifiers.Finally,by using the way of cascading to combine haar-like feature and texture feature to build an intrusion recognition classifier,which can accurately realize classification and recognition of intrusion object,and can be divided into intrusion foreign object and non intrusion foreign object.Through the experimental simulation of the collected video,the results show that: in this thesis,the machine vision based railway track foreign object intrusion detection algorithm fully considers the characteristics of the railway scene,and has high detection accuracy during the processing.