基于GPU并行运算的实时定位方法研究OA

To meet real-time and high-precision localization requirements in rail transit scenarios,a parallel computing localization method based on LiDAR and inertial measurement units(IMU)is proposed.Adopting a space-for-time design concept,this method utilizes the GPU characteristics:a large number of cores,strong parallel computing performance,and high video memory bandwidth.By constructing a voxel grid network,it enables efficient spatial indexing and nearest neighbor search of point clouds;voxel downsampling,LiDAR point cloud preprocessing,line and surface constraint construction,localization initialization,and relocation modules are transplanted to a graphics processing unit(GPU)for parallel operations.Experimental results show that the proposed method meets real-time localization requirements for automatic train operation from both localization accuracy and real-time performance,with an average decrease in CPU load rate by 25.66%,a 44.72%reduction in single-frame point cloud processing time from 85.87 ms to 47.47 ms,and an average absolute pose error of 0.10 m.