Automatic fault recognition for Brake-Shoe-Key losing of freight train

Based on the shape descriptors of image contour, this paper proposes an automatic fault recognition method for Brake-Shoe-Key (BSK) losing of freight train. The method contained two phase, one is the positioning of the feature region, the other is recognition of the feature for the BSK losing. In the position phase, a direction-adaptive grey projection method followed by an enhancing process was put forwarded for the initial positioning of image feature region of the Brake Shoe. And then based on edge information, a threshold-relaxation method was adopted to precisely extract the contour of the Braked Shoe. In the recognition phase, besides some ordinary shape descriptors such as rectangular degree, solidity, compactness, circularity, etc., two new shape descriptors, respectively named as contour smoothness and contour concave-convex were proposed to establish the recognition characters vector. At last, the Support Vector Machine (SVM) was used in the classifier for BSK losing fault recognition. Experiments demonstrated that the proposed algorithm is not sensitive to image noise and performed well on images with complex background and disturbance such as blur, poor illumination, excess exposure, etc.