A vision-based serial number recognition algorithm for HSR trains by nearest neighbor chains of connected components

With the rapid development of High Speed Railway (HSR) in recent years, its research become one hot academic topic. Serial numbers of HSR trains are unique identifications, which play an important role in railway management and operation. In this paper, we present a vision-based algorithm to automatically recognize the serial numbers of HSR trains by image sensors. Firstly, according to the fixed character layout of serial numbers, the serial number regions in the image are located by a combination of connected components. Maximally Stable Extremal Region (MSER) detector is introduced to extract reliable connected components as candidate characters. Based on the pairwise geometry relation between candidates, these connected components constitute the nearest neighbor chain. The accurate location of the serial number is obtained with the nearest neighbor chain. Meanwhile, character images can be segmented simultaneously. Finally, the characters are recognized with Histogram of Gradient (HoG) features and simple similarity-based classifiers. In the experiments, our recognition performance is evaluated by the test images which are collected from real application scenes. The experimental results show the reliability and effectiveness of our serial number recognition algorithm.