Research And Implementation Of Disease Inspection For Track Connector Based On Machine Vision

With the rapid development of rail transit,the maintenance of rail lines is becoming more and more important.Rail fasteners,rail welds and other rail connectors as an important part of the track,mainly to achieve the rail and sleeper connection and rail location fixed.The traditional method of manual detection has been unable to meet the requirement of track intelligent detection.In addition,due to the harsh environment of subway tunnels,the safety and health problems of detection workers cannot be guaranteed.Therefore,intelligent detection of track lines has become an important development direction in the future.The machine vision inspection system of track connector disease studied in this paper,as a key part of track connector disease inspection robot,provides a reliable and efficient detection method for intelligent and lightweight detection of track connector.Aiming at the problem of track connector inspection such as track fastener defect and rail weld abnormality,this paper designed and implemented an adaptive machine vision inspection system for rail connectors based on the track connector disease inspection robot platform independently developed by our research group.The main work content as follows:Firstly,through in-depth investigation of the status quo of subway track inspection,understand the inspection requirements of track connectors and establish the track fastener and rail weld detection targets.Based on this,the overall structure of the machine vision inspection system for the rail connector inspection is designed and constructed,and the whole system is designed and implemented in a modular way.The second is to clarify the various modules of the machine vision inspection system,which mainly includes the light source and its control module,image acquisition and transmission management module,image processing and algorithm module,and design and implement each module.For the light source and its control module,it is designed and implemented through the specific work of the hardware structure selection design,optical characteristic analysis and verification,design and implementation of the light source control unit and test analysis to ensure that the camera can collect images with uniform brightness and prominent object features.As for the image acquisition and transmission management module,combined with the characteristics of the track connector disease inspection robot platform and the detection requirements,the analysis and selection of key hardware such as camera,lens,lower computer processor and cloud server.The camera and lens parameters are debugged through multiple tests,and the external trigger of the camera is realized.At the same time,the semi-reference method is used to evaluate the sharpness of the image to ensure high quality images are captured.Then use the 4G mobile network to transmit the image that meets the requirements and the image processing detection result to the cloud server to realize the data storage and management.The image processing and algorithm module is mainly used to study the algorithm of the track connector detection process.Firstly,the original image is denoised and enhanced,and then the position of the track connector in the image is located to obtain the region of interest.The gradient direction histogram features and local binary pattern features of the region of interest are extracted and merged into HOG-LBP features.Finally,the track connector sample set is constructed by SVM classification algorithm for offline training,and the SVM model of track connector state detection classification is obtained.Finally,the integration test of the system.The various modules of the track connector disease machine vision inspection system studied in this paper are integrated into the track connector disease inspection robot platform for testing,and the results are analyzed to verify the accuracy and feasibility of the inspection algorithm of the system.