Research And Implementation On Detection Algorithm Of The Rail Surface Defects Based On DSP

With the rapid development of railway transportation, people pay more and more attention to conduct safety inspections of the track components. For a long time,our country in this area is dependence on mass experience workers for inspection, not only time-consuming, low degree of automation, and high-speed operation of the train will be a great threat to the personal safety of Patrol Road workers. The machine vision system has the advantages of speed, high precision, non-contact etc, and is applied to the rail surface defects detection, can effectively overcome the deficiency of the manual inspection,to improve the degree of automation of the security screening equipment rail parts. And Visual inspection system to complete the online detection of the rail surface defects, must raise the visual image processing speed, DSP can well meet the requirements, DSP with its incomparable speed in the video intelligent monitoring, multimedia applications and other fields have a wide range of applications, in the visual inspection as a new independent applications also more and more cause researchers’attention. In this context, combined with the advantages of embedded system, this paper focused on the rail surface defects detection system of special applications, study the rails defects detection algorithms,and use the algorithm for embedded hardware implementation by DSP.In this paper, focusing on the following aspects of research:Puts forward a new method of rail regional orientation, namely according to many times concluded that the actual value of the static to determine the cutting area boundary parameter, and cut its orbital zone localization, this method not only can reach the effect of regional orientation to orbit, and only in software program directly the specific cutting parameters have laboratory values to set parameters complete cutting, save for the orbit regional orientation of the operation, simplify the whole image processing algorithm process, compared with other positioning method, saving time.Put forward based on the improved adaptive Canny algorithm track surface defects detection algorithm. Through to the commonly used method of image segmentation analysis and experiment, found that based on the edge of the image segmentation method is suitable for the fast track surface defects detection requirements. And in the light of the Canny algorithm smooth scale parameters and double adaptive threshold value of the poor is improved, and puts forward a new Canny algorithm to complete rail defect detection. The algorithm of pixel grayscale value and filter in the window in the difference image grey value as a gaussian filter parameters, is used to measure the parameters to adjust gaussian filter method instead of the fixed scale parameter of the gaussian filter method, better according to different complexity to achieve real image noise removing. With the improvement of the double threshold adaptive selection method of double replacing fixed threshold selection method, that is, based on the specific image characteristics, to determine each image different threshold thus up to double the adaptive threshold value choice.According to the defects of the main geometric feature extraction and shape characteristic, analyzes the selected shows the difference between two kinds of defects aspect ratio and complexity as the input characteristic vector classifier, Designed for this topic research object of LVQ neural network classifier scar and crack two types of defects accurate classification.According to the project application characteristics and the current trend, choose the synthesis processing ability of TI special video processing DSP-DM642, and built based on DSP embedded machine vision system image processing platform, complete hardware of initialization and commissioning work.Completed the algorithm DSP transplantation work, the first in Matlab7.0environment of the algorithm simulation and improve research, and then in the TI provide software development environment CCS3.1DSP developed in the program, completed the algorithm realization of hardware development and DSP, that is, in the final image processing DSP platform,Completed image collection,rail regional orientation, smooth,edge detection and identification.