Research On Identification Methods Of Track Defects Based On Machine Vision

Track detection is the important guarantee for the safety the track working. Facing the current situation of the rapid development of railway enterprise in our country, the existing methods have been unable to meet the requests. Machine vision is an effective method to long distance, non-contact, automated detection. Thus, the research of the vision inspection technology of track defects and the portable detection car of track defects on machine vision is achieved,which realizes the digital management of the information of the rail and fastener defects.Firstly, basing on the basic design idea of machine vision and the functional requirement of track detection, the system of portable track defect visual inspection is designed. The hardware designs, including the system acquisition scheme, equipment selection and so on, are completed by the combination of the track design standards and the principle of camera imaging. According the basic methods of image processing, the orbital defect detection algorithm and other system software designs are completed.Next, track regions of hue value characteristics of each part are analyzed. Using the hue value mutation characteristics of the edge of the rail surface, combining with the feature of rail shape, fitting the linear of rail boundary points, the rail surface area is extracted. Similarly the discriminant of sleeper region is extracted. Combing the characteristics of the cross with rail and sleeper and the size of fastener area, the fastener region is extracted.Then, gray level compensation and adaptive weighted median filtering for the image of the rail surface area is used to enhance the defect characteristics. To solve the problem of burr and broken edge in the image of binarization, the morphological processing of the image based on the travel encoding method is optimized. The defect region is determined by contour tracking based on chain code direction. The geometric features and shape features of the defects are recorded, and the simple classification of Scar and crack are made by using the ratio of length-width.Moreover, the switching median filter is used to remove the impulse noise which has a serious influence on the edge of the fastener. Histogram concavity feature image gradient magnitude solves the problem of double threshold adaptive selection in Canny operator edge extraction. According to the curve features of the stability of the missile, the template matching based on the curve projection is carried out, the detection of loss has been realized rapidly.Finally, the portable track defect system with visual inspection is written by LabVIEW. The field experiment is carried out on a portable track defect vehicle of the image detection based on machine vision. Through the test of the target area extraction under different light intensities, the system shows that the system has the ability to resist the interference of light intensity. The effective detections of the rail surface defect and the lack of fasteners detection are verified by using the image of the collected images and some of the defecting images of synthesis. Meanwhile the detection time meeting the real-time detection requirements, the system can replace manual inspection in a certain extent.