The millimeter-wave radar has good weather robustness, but currently lacks performance in object classification. With the advent of imaging radar, three-dimensional (3D) point clouds of objects can be obtained. Based on 3D radar point clouds, an support vector machine (SVM) algorithm using 3D features is proposed to solve poor radar classification performance. First, a new 29-feature vector is proposed from many perspectives, such as shape features, statistical features, and other features. Then the SVM classifier with four different kernel functions and other machine learning methods are used to achieve multi-objective classification. Finally, experiments are carried out on three types of datasets collected by ourselves, and the results show that the algorithm achieves a 95.1% classification accuracy, which is 15.7% higher than the traditional 2D radar point cloud. Moreover, PointNet and PointNet ++ networks are firstly used in the 3D radar point clouds and compared with the SVM algorithm, showing the feasibility of deep learning in radar 3D point clouds. It is an important reference for improving the accuracy of radar multi-objective classification algorithms and migrating deep learning to radar 3D point clouds.
Multi-Objective Classification of Three-Dimensional Imaging Radar Point Clouds: Support Vector Machine and PointNet
Sae Intl. J Cav
Sae International Journal of Connected and Automated Vehicles ; 4 , 4 ; 371-382
2021-10-21
12 pages
Conference paper
English
Segmentation of Drone Collision Hazards in Airborne RADAR Point Clouds Using PointNet
ArXiv | 2023
|HIGH DIMENSIONAL FRUSTUM POINTNET FOR 3D OBJECT DETECTION FROM CAMERA, LIDAR, AND RADAR
British Library Conference Proceedings | 2020
|