基于特征提取改进的LeGO-LOAM方法

Improved LeGO-LOAM Method Based on Feature Extraction

针对LeGO-LOAM算法中的局限性,提出一种基于特征提取改进的LeGO-LOAM方法.由于室外复杂环境下随机特征点较多,首先采用自适应特征提取方法在不同的距离下保证特征提取的准确性;然后基于随机一致性采样优化的地面滤除方法去除不可靠特征点;最后通过相邻帧中同一块特征点的移动距离来消除动态目标.利用KITTI数据集进行仿真分析,结果表明本文优化后的平均误差降低了19.1%,最大误差降低了23.9%.在实际激光雷达建图中进行多组试验,计算出的位置最大误差小于1m,表现出了良好的鲁棒性和稳定性,优化效果显著.

An improved LeGO-LOAM method based on feature extraction is proposed to address the limitations in the LeGO-LOAM algorithm for lightweight LiDAR mapping.Due to the large number of random feature points in the complex outdoor environment,an adaptive feature extraction method is firstly used to ensure the accuracy of feature extraction at varying distances,followed by a ground filtering method based on optimized random consensus sampling to remove unreliable feature points.Finally,dynamic targets are eliminated by measuring the moving distance of the same set of feature points across adjacent frames.Simulation analysis using the KITTI dataset shows that the optimization in this work result in an average error reduction of 19.1%and a maximum error reduction of 23.9%.Through multiple sets of experiments conducted in actual LiDAR mapping,the maximum error of the calculated position is less than Im,showing good robustness and stability with remarkable results.