Railway Fastener Defects Detection Using Gaussian Mixture Deformable Part Model

This paper addressed the problem of detecting the completely missing and partly missing railway fasteners in the collected images. A Gaussian mixture deformable part model (GMDPM) algorithm was proposed using histogram of oriented gradient (HOG) features. The fastener template was divided into four parts considering the shape of the fastener, and seed points were uniformly sampled along the fastener’s shape contour. The part and the seed point deformation were defined to fit the deformation of the fastener. Each seed point template in the part model was solved iteratively by using Gaussian mixture model (GMM) with an expectation-maximization algorithm. The results reveal that the proposed method achieves good performance, especially when the illumination difference is large and the fastener is partially occluded or has slight shape deformation.