Application of Train Intelligent Detection System in Large-Volume Operated Lines

As one of the most important public transportation, urban rail transit bears great operational pressure and relies completely on signaling system to maintain its safety and efficiency. In order to improve train operation efficiency and safety, train active obstacle perception is a good solution. In this paper, we introduce the train intelligent detection system (TIDS) and its application in practical line. TIDS consists of data pre-processing layer, perception layer, and decision-making layer. The data pre-processing layer mainly includes data acquisition and spatio-temporal synchronization. The perception layer applied convolutional neural network to achieve rail area detection and train detection simultaneously in each camera image, and further combine lidar points to measure the train distance. And the decision-making layer judge the relationship between detected trains and the rail area, so as to determine whether the detected trains affect the normal running of the train. At this stage, TIDS system has been comprehensively tested on Beijing Yanfang Line, Shanghai Line 6, Chengdu Line 3, Hong Kong Tsuen Wan, etc. It has been tested and accumulated more than 2 TB data. In addition, we analyze the running accuracy of TIDS system in detail, the results showed that the leak detected rate (LDR) is 0.007 and the mistakenly detected rate (MDR) is 0.000478. Particularly, the system is being practical applied in Hong Kong Tsuen Wan line.