用于移动机器人导航的通用激光雷达特征提取

General purpose LIDAR feature extractor for mobile robot navigation

提出了一种基于结构张量的、无需先验知识的、可广泛应用于各种环境的多尺度激光雷达数据特征提取方法.这种方法与经典的图像处理方法具有相似的数学意义,利用数据对应的结构张量来评价特征的显著性.在经典数据集上对比了文中提出方法与经典方法的重复度和计算复杂度,实验结果表明:这种方法具有广泛适用性,并比传统方法具有更高的精度和重复度.

A structure tensors based,prior knowledge independent,general purposed feature extractor for LIDAR(light detection and ranging)data was proposed.The method shared ts mathematical foundation with classical visual feature detectors,and they measured strength of features based on structure tensors.The precision and the repeatability of classical methods were compared with the proposed one on representative datasets.Experimental results show that the proposed method does not depend on prior knowledge,and it outperforms classical methods on both precision and repeatability.