A fast template matching-based algorithm for railway bolts detection

Railway bolts detection is an important task in railway maintenance and some techniques based on traditional feature extraction and classification have been used in this application. However, these techniques have two critical disadvantages, i.e., manual collection of training data set and time-consuming training process; furthermore, trained classifiers are hard to generalize from a specific railway to the others. In order to overcome these problems, we propose a fast template matching-based algorithm, named FTM, in this paper. Firstly, we use a template matching method to locate the bolts with constrains of the railway geometric structure. Then, we use a nearest neighbor classifier to determine whether a bolt is in position or not. At last, we use GPU with CUDA architecture to accelerate the most time-consuming part of FTM. The experiments demonstrate that our proposed FTM algorithm achieves the accuracy of 98.57% in average, and the average false positive is only 0.89%. The overall speedup of FTM by GPU is 6.11, and the most time-consuming part gets speedup of 17.73. Furthermore, FTM only need to collect several samples in a new railway without laborious training work.