列车自主定位图像识别技术应用研究

Application of Image Recognition Technology for Train Autonomous Positioning

为进一步提高新型列车运行控制系统列车自主定位技术定位精确度及地形适应能力,解决基于北斗卫星的列车定位技术方案具有的卫星信号“盲区”问题,分析图像识别技术需求,阐述图像识别技术方案。在此基础上,结合运营场景,分析运行状态综合判断、图像预处理、视觉标签定位、视觉标签识别、位置计算及报告等模块的功能,以及各软件模块间如何协同工作。构建深度神经网络模型进行模块功能实现,并使用大量训练数据集对其进行训练及仿真验证,得出新设计的列车自主定位图像识别技术方案的准确率达到93.25%,F-Measure评价值达到89.99%,为列车自主定位技术提供了新的参考依据。

In order to improve the positioning accuracy and terrain adaptability of the new train autonomous positioning technology of the new train control system,and to solve the satellite signal problem of range for the Beidou satellite-based train positioning technical solution,this paper analyzes the requirements of image recognition technology and explains the image recognition technology plan.On this basis,combined with operating scenarios,this paper analyzes the functions of modules such as comprehensive judgment of operating status,image preprocessing,visual label positioning,visual label recognition,position calculation and reporting,and the collaboration of different software modules.A deep neural network model is built to realize the module function,and a large number of training data sets are used for training and simulation verification.It is concluded that the accuracy of the newly designed train autonomous positioning image recognition technology scheme reaches 93.25%and the F-Measure evaluation value reaches 89.99%,providing a new reference basis for train autonomous positioning technology.