Millimeter Wave Point Cloud Processing Method Based on Density Clustering

Abstract

With the continuous improvement of millimeter wave radar hardware performance, radar point cloud, as an important imaging method for millimeter wave radar, has been widely used in mobile robots, autonomous driving, and environmental perception tasks. Compared with conventional LiDAR and visual sensors, millimeter wave radar has significant advantages in all-weather, avoiding privacy leaks, and has reliable stability in complex climate scenes. However, its measurement accuracy and relatively small amount of data are drawbacks, and there are a large number of outliers and other noise data in the point cloud data obtained from the surrounding environment. Aiming at the problem of noise point filtering, an improved density clustering (DBSCAN) algorithm is proposed. By traversing the sliding block to find the densest region and finding the initial core point, the clustering radius is adaptively adjusted according to the sparsity of the point cloud within the initial radius to complete the clustering filtering of noise points. By comparing the experimental results, the improved clustering algorithm has better accuracy and reliability in obtaining point cloud in various different scenarios, and the quality of point clouds has been improved.