Detection Method of Track Locating Point Based on Yolo V3

This paper presents a method to determine the position of the train in the track by detecting the track locating points, and establishes the image data sets of the track locating points. Because the locating points on each track are unique, the number of locating point image samples is very small, which poses a great challenge to the accuracy of locating point detection. We apply the target detection algorithm of Polo V3 to the field of track location point detection, and propose three improvements. Firstly, the training data sets are expanded by data enhancement of images. Then K-means clustering algorithm is used to analyze the size of the anchor boxes of the data sets, and new clustering centers are obtained. Finally, the multi-scale training method is used to make the model adapt to images of different resolutions. The results indicate that compared with the original network, the improved Yolo V3 model not only has better adaptability to image detection with different quality and resolution, but also has higher mean average precision and better detection effect.