Identification of High-speed Railway Trackside Equipments Based on YOLOv4
Sian Qian, Shenghua Dai · 2022
In view of the shortcomings of low detection accuracy and low efficiency in the traditional manual detection of trackside equipment by using simple measurement tools, this paper aims to obtain an efficient and fast detection method for identifying all trackside equipments through model training based on YOLOv4 target detection algorithm and taking the balise, telephone pole and telephone pole number on highspeed railway line as the research object. In the process, the identification method of multi-target trackside equipments is obtained through the steps of labeling the original data set, enhancing data set, obtaining the anchor box by improved k-means clustering, training and tuning YOLOv4 model, and the detection scheme is evaluated by comparison. The results show that after improving the anchor box parameters, the model accuracy improved by 7.23%, the value can reach 84.83%, the video detection speed is also increased by 7.58%, and the detection performance is significantly improved.