基于分块地图和点云配准的智能车辆定位方法OA

To address the issue of low localization accuracy of intelligent vehicles due to satellite signal attenuation in complex urban driving scenarios,an intelligent vehicle localization method based on block maps and point cloud registration is proposed.Firstly,a tightly-coupled LiDAR-inertial odometry framework is applied to construct a high-precision map of the vehicle's driving environment,capturing prior environmental information of the scene.Then,a block map is generated by deploying a set of keyframes,and a dynamic sliding-window switching strategy is used for boundary management.Additionally,a branch-and-bound search strategy is introduced to determine the loading area of the prior map.The Scan Context descriptor is then utilized for matching to determine the initial pose of the vehicle within the regional map.Subsequently,the fast truncated least squares estimation and graduated non-convexity(GNC)point cloud registration algorithm are employed to iteratively solve for the precise pose based on the obtained initial pose.Finally,the proposed localization method was evaluated on both public datasets and real vehicle platforms.The test results in urban park scenarios demonstrate that the proposed global localization method achieves a good balance between time efficiency and accuracy,with a global localization error of 0.14 meters and a localization time of 0.76 second.Compared to existing localization algorithms,the maximum error is reduced by 31％,and the average lateral error of the vehicle during road driving is reduced to 0.15 meters.