Research And Implementation Of Key Technology For Locomotive Wheel Image Matching

As one of the most important units to keep the safety of train, running gear has always been the focusing point of scholars at home and abroad. And amount of running monitoring systems has been developed to monitor the stability of running gear on the train, when the trains are running on the rail. While running gear consists of many components such as wheel etc., which are needed to be evaluated by manual percussion. Because of low efficiency and after judgement of manual percussion, the technique of image matching is used to match wheel image in this thesis with machine vision, which is able to improve the efficiency of train maintenance and has a reference value to keep the safety of running train.Based on image matching technology, the key techniques of wheel image matching are studied, and a wayside image capture system is established. By using of constant light source and fixed camera, this image capture system is characterized with illumination invariant, scale invariant and rotational invariance. Taking standard wheel image as a template image, the captured image is matched with template matching method. The experimental results show the correlation of matched image is higher than 0.9, and the matching accuracy is up to the pixel level.In the meanwhile, the detail steps of SIFT algorithm is studied, and SIFT algorithm based wheel image matching model is established. With the analysis of wheel images’ features, the threshold limiting the edge points is weakened to make sure more feature information can be extracted. At last, the threshold limiting parameter is set to 10 with thorough experiments. Followed, on account of wrong matching of feature points, the reverse matching algorithm is mentioned to improve the original matching algorithm. And the reverse matching algorithm can control the positive matching result to eliminate lots of wrong matchings and improve the recall, which can be improved up to 100%. At last, Similarity measurement algorithm is analyzed. Euclidean distance is replaced with the linear combination of Street distance and chess distance to lower the complex of calculation and improve the efficiency of original SIFT algorithm by 40%. At last, the improved SIFT algorithm is used in wheel image matching.Simulation and experiment results show that the improved SIFT algorithm can improve the recall and efficiency of original SIFT algorithm, and keep the invariant of original SIFT algorithm in the mean while. Also the precision is improved to sub-pixel, which has a certain reference value and vital significance in the train component targeting.