基于多源融合的智能列车同时定位与侵限检测方法研究

Research on Simultaneous Localization and Intrusion Detection Methodology for Intelligent Trains Based on Multisource Fusion

为解决智能列车运行时的自主定位和碰撞预警问题,文章提出一种基于多源融合的智能列车同时定位与侵限检测方法。其首先利用激光雷达、惯性测量单元(IMU)和点云地图数据,构建最小二乘优化模型,并基于图优化理论解算列车运动状态,完成列车定位;接着,基于去畸变点云、列车定位信息和地图先验数据,构建列车运行的实时3D限界,并基于深度图分割和点云聚类算法完成限界内的目标检测,获取目标的位置和尺寸信息;然后,采用视觉检测技术从图像中获取目标的类别信息;最后,基于列车定位信息和IMU数据,完成激光雷达数据与视觉数据的时间同步,将激光雷达检测到的目标与视觉检测到的目标在图像平面内进行融合,得到目标的类别、位置和尺寸信息,完成侵限检测。实验结果表明,采用该方法,列车的平均水平定位偏差不超过20cm,限界内止挡的检测准确率达到97.91%。该方法可实现精确、鲁棒的列车定位和侵限检测功能。

To achieve self-localization and collision warning for intelligent trains,a novel methodology for simultaneous localization and intrusion detection based on multisource fusion is proposed.The initial integration of LiDAR,inertial measurement unit(IMU),and point-cloud map data facilitates the development of a least squares optimization model,and this model is used to determine train motion states based on the graph optimization theory,fulfilling the task of train localization.Distortion-less point-cloud data,train localization data,and prior map data are then fused to create real-time 3D clearances for train operation.Following this,object detection is performed within these clearances based on depth map segmentation and a point cloud clustering algorithm to acquire the positions and sizes of intrusion objects.Subsequently,a visual inspection technique is applied to classify objects from images.Finally,time synchronization is established between the LiDAR and visual data,based on train localization information and IMU data.This allows for the fusion of objects detected by LiDAR and those from visual detection within image planes,yielding the categories,locations,and sizes of those detected intrusion objects at the conclusion of the intrusion detection process.Experimental results demonstrated deviations of not more than 20 cm in the transverse localization of trains and detection accuracy of up to 97.91%for buffer stops within clearance ranges.The proposed approach provides accurate and robust results of both train localization and intrusion detection.