| With the development of high-speed railway, the proportion of EMU in the national railway trains has increased a lot, and people are paying more and more attention to the security of EMU. The structure of EMU is complex with various components. Currently, there is a system called running Trouble of Freight Car Detection System(TFDS) which is widely used to maintain the safe operation of the train. However, it can only detect some key parts of the train, it can't detect the whole components of the EMU. Therefore, Trouble of moving EMU Detection System(TEDS) has been put forward.This paper studies EMU image matching and the abnormal region recognition in the TEDS. To locate the feature area, it uses the sampling rate and speed of the vehicle to estimate the area, which can narrow the searching range of image matching and improve the reliability of positioning. In the image matching procedure, for feature areas which contain significant linear edges of the structure, it proposes a significant edge extracted algorithm by using a line merging idea, which can improve the robustness when the lights change. For feature areas with complex edge structure, it proposes a significant area selection method based on the peak point of the edge projector image, then use the edge point distance transform algorithm to quantify the similarity measure, and devise an algorithm to accomplish the feature area matching, the algorithm improves the real-time performance and adaptability. To detect the abnormal region of the registered images, it proposes algorithm based on vote by local window, which can eliminate the influence of train shaking and incomplete registration, and improve the applicability of the algorithm.Finally, the algorithm above were simulation tested based on a n image library, positioning ratio, matching accuracy, abnormal recognition rate and the error rate of those algorithms can all meet the demand, and those algorithm are proved to be effective with an reasonable running time. |
PDF Full Text Request