基于阈值的激光雷达K均值聚类算法

针对同一距离不同目标的激光雷达全波形回波数据聚类准确率低的问题，在分析K均值聚类算法原理的基础上，提出了一种基于阈值的K均值聚类算法。首先，利用强度信息对距离信息进行标定，使用强度信息作为特征进行聚类以区分同距离的不同目标。然后，利用阈值限定聚类中心间的最小距离，提高聚类准确率。最后，搭建了扫描验证平台进行平移和旋转成像，对算法有效性进行验证。通过不同颜色目标和模拟道路回波数据聚类实验表明，在不同阈值的情况下，提出的基于阈值的K均值聚类算法的聚类准确率均在90%以上，相比于无阈值的K均值聚类算法准确率提升10%以上，能够有效进行目标聚类和模拟道路提取。 

LiDAR K-means clustering algorithm based on threshold

For this question of low clustering accuracy problem in LiDAR full-waveform echo data with different targets at the same distance, a threshold-based K-means clustering algorithm was proposed based on the analysis of K-means clustering algorithm. Firstly, The distance information was calibrated using the intensity information, and the intensity information was used as a feature to distinguish different targets at the same distance by clustering. Secondly, the threshold was used to define the minimum distance between clustering centers to improve the clustering accuracy. Finally, the scanning verification platform was built for translation and rotation imaging to verify the effectiveness of the algorithm. The clustering experiments of different color targets and simulated road echo data show that the clustering accuracy rate of threshold-based K-means clustering algorithm is above 90% under different thresholds, and increases more than 10% compared with the threshold-free K-means clustering algorithm, which can effectively perform target clustering and simulation road extraction.