基于视频的轨道车辆自主定位方法研究

Vision-Based Absolute Position Extraction Method for Rail Vehicles

摘要:

针对轨道施工车辆自主定位需求，提出一种基于车载前视相机拍摄百米标视频的轨道车辆自主定位方法。该方法首先对YOLOX-s网络进行改进并构建了百米标的目标检测模型，完成对百米标的目标检测；其次，当检测到百米标后，结合图像预处理和卷积循环神经网络（CRNN）网络构建百米标数字文本识别模型，提取百米标的数字文本信息，从而实现对轨道施工车辆的定位。经实验验证该方法能够快速准确定位轨道施工车辆的位置信息。

关键词:轨道车辆定位;机器视觉;目标检测;文本识别

Abstract:

Aiming at the demand of autonomous positioning of rail construction vehicles， this paper proposes a method of absolute position extraction of rail vehicles based on the video of 100-metre markers captured by the on-board forward-looking camera. The method first improves the YOLOX-s network and constructs the target detection model of the 100-metre marker to complete the target detection of the 100-metre marker. Then， when the 100-metre marker is detected， it combines image preprocessing with convolutional recurrent neural network （CRNN） network to construct the 100-metre marker digital text recognition model to extract the digital text information of the 100-metre marker， so as to achieve the absolute position location of rail construction vehicles. The method is verified to be able to quickly and accurately extract the absolute position information of rail construction vehicles.

Key words:rail vehicle positioning;machine vision;target detection;text recognition