基于激光雷达的远距离运动车辆位姿估计
发布时间：2018-07-25 08:24
【摘要】：为了解决激光雷达扫描远距离运动车辆产生的点云稀疏导致位姿特征难以提取的问题,提出了一种远距离运动车辆位姿估计方法。首先利用时空连续性提取远距离运动车辆。然后利用最小二乘拟合得到稀疏点云水平面二维投影近似拟合直线对,依次在不同角度的垂直正交直线对上对稀疏点云的二维投影进行一维向量估计的装箱过程,基于目标车辆与激光雷达间相对位置的观测角函数最大化匹配滤波响应,进而利用全局优化算法对投影点概率分布与匹配滤波运算得到的代价函数作离散卷积,寻优比较得到单帧拟合最优矩形。最后结合连续帧平移约束进行多帧拟合,优化当前帧目标车辆拟合矩形的位姿。利用仿真和真实场景下采集的目标车辆点云数据进行算法验证分析。结果表明:在点云稀疏的情况下,当远距离目标车辆做直线运动时,提出的多帧拟合方法得到的位姿参数均方根误差低于单帧拟合和已有的RANSAC拟合方法;当远距离目标车辆做曲线运动时,提出的单帧拟合和多帧拟合方法得到的位姿估计结果较为接近,且误差明显低于已有的RANSAC拟合方法;对于不同相对距离下采集的目标车辆点云,提出的单帧拟合和多帧拟合位姿估计方法的适应性优于已有的RANSAC拟合方法。
[Abstract]:In order to solve the problem that the point cloud sparsity caused by the remote moving vehicle scanned by lidar makes it difficult to extract the pose feature, a method of position and attitude estimation for long range moving vehicle is proposed. First, the spatiotemporal continuity is used to extract the long distance moving vehicle. Then the least square fitting is used to get the approximate fitting line pair of sparse point cloud plane two-dimensional projection, and the packing process of one dimensional vector estimation of the two-dimensional projection of sparse point cloud is carried out on the vertical orthogonal straight line pairs of different angles in turn. Based on the maximum matched filtering response based on the relative position between the vehicle and the lidar, the global optimization algorithm is applied to the discrete convolution of the probability distribution of the projection point and the cost function obtained by the matched filtering operation. Single frame fitting optimal rectangle is obtained by optimization comparison. Finally, multiple frames are fitted with successive frame translation constraints to optimize the position and orientation of the current frame target vehicle fitting rectangle. The algorithm is verified and analyzed by using the point cloud data of the target vehicle collected in the simulation and real scene. The results show that the root mean square error of pose parameters obtained by the multi-frame fitting method is lower than that of single frame fitting method and RANSAC fitting method when the vehicle is moving in a straight line when the point cloud is sparse. When the long distance target vehicle moves on the curve, the position and pose estimation results obtained by the single frame fitting method and the multi frame fitting method are close, and the error is obviously lower than that of the existing RANSAC fitting method. For the point clouds of target vehicles collected at different relative distances, the proposed methods of single-frame fitting and multi-frame fitting are more adaptable than the existing RANSAC fitting methods.
【作者单位】： 长安大学汽车学院;西安工业大学机电工程学院;
【基金】：国家自然科学基金项目(61473046) 中央高校基本科研业务费专项资金项目(310822151028,310822172001) 陕西省自然科学基金项目(2016JQ5096) 长江学者和创新团队发展计划项目(IRT1286)
【分类号】：TN958.98;U463.67
本文编号：2143212
[Abstract]:In order to solve the problem that the point cloud sparsity caused by the remote moving vehicle scanned by lidar makes it difficult to extract the pose feature, a method of position and attitude estimation for long range moving vehicle is proposed. First, the spatiotemporal continuity is used to extract the long distance moving vehicle. Then the least square fitting is used to get the approximate fitting line pair of sparse point cloud plane two-dimensional projection, and the packing process of one dimensional vector estimation of the two-dimensional projection of sparse point cloud is carried out on the vertical orthogonal straight line pairs of different angles in turn. Based on the maximum matched filtering response based on the relative position between the vehicle and the lidar, the global optimization algorithm is applied to the discrete convolution of the probability distribution of the projection point and the cost function obtained by the matched filtering operation. Single frame fitting optimal rectangle is obtained by optimization comparison. Finally, multiple frames are fitted with successive frame translation constraints to optimize the position and orientation of the current frame target vehicle fitting rectangle. The algorithm is verified and analyzed by using the point cloud data of the target vehicle collected in the simulation and real scene. The results show that the root mean square error of pose parameters obtained by the multi-frame fitting method is lower than that of single frame fitting method and RANSAC fitting method when the vehicle is moving in a straight line when the point cloud is sparse. When the long distance target vehicle moves on the curve, the position and pose estimation results obtained by the single frame fitting method and the multi frame fitting method are close, and the error is obviously lower than that of the existing RANSAC fitting method. For the point clouds of target vehicles collected at different relative distances, the proposed methods of single-frame fitting and multi-frame fitting are more adaptable than the existing RANSAC fitting methods.
【作者单位】： 长安大学汽车学院;西安工业大学机电工程学院;
【基金】：国家自然科学基金项目(61473046) 中央高校基本科研业务费专项资金项目(310822151028,310822172001) 陕西省自然科学基金项目(2016JQ5096) 长江学者和创新团队发展计划项目(IRT1286)
【分类号】：TN958.98;U463.67
【相似文献】
相关期刊论文 前2条
1 罗勇;陈慧;;基于EKF的自动泊车系统位姿估计算法设计[J];山东交通学院学报;2013年04期
2 孙团;王艳华;吴猛猛;;发动机活塞的测量与点云处理[J];科技视界;2012年09期
，本文编号：2143212
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2143212.html
