基于最近迭代点的毫米波雷达点云数据处理方法

毫米波雷达具有小型化、低成本等特点，可全天候、全天时工作，在高级辅助驾驶系统中发挥着重要作用。基于多芯片级联方案的多发多收（multiple-input multiple-out, MIMO）技术可有效提高毫米波雷达的角度分辨率，使得毫米波雷达点云成像成为可能。针对毫米波雷达图像点云稀疏、噪点多等问题，本文提出了一种基于最近迭代点的毫米波雷达多帧融合和自适应邻域半径的DBSCAN算法。首先，利用MIMO毫米波雷达技术获得多帧观测场景目标点云图像。其次，利用辅助信息得到点云匹配的初值，通过最近迭代点算法估计平移旋转矩阵进行精确匹配，实现多帧数据融合以改善图像点云稀疏问题。然后，设计自适应阈值的DBSCAN算法去除噪声，获得目标的点簇信息，再对聚类后的点簇目标求取最小外接矩形，结合目标散射强度，实现对车辆和围栏等不同类型目标区分。最后，利用外场（典型停车场场景）测试数据，对本文所提算法的有效性进行了验证。

Iterative Closest Point Method for Point Cloud Data Processing of Millimeter Wave Radar

? ?As an all weather and all day sensing tool， millimeter-wave radar with the characteristic of miniaturized size and low cost plays an important role in advanced assisted driving systems. The multiple-input multiple-out （MIMO） technology based on multi-chip cascade scheme can effectively improve the angular resolution of the millimeter-wave radar， making it possible to provide point cloud images of scene. To handle the drawbacks of sparse point cloud and serious noise in millimeter-wave radar image， this paper proposes a novel algorithm of multi-frame fusion based on iterative closest point （ICP） processing and density-based spatial clustering of applications with noise （DBSCAN） which has adaptive neighborhood radius. First， MIMO millimeter-wave radar technology is employed to obtain the point cloud image of the target in the observation scene. Secondly， the odometry information is used to obtain the initial value of point cloud matching， and then the iterative closest point algorithm is used to estimate the translation and rotation matrix to achieve accurate matching. In this way， the problem of sparse point cloud is effectively solved through multi-frame data fusion. Then， the noise is removed by the adaptive threshold DBSCAN algorithm to obtain the point cluster information of the target. The minimum circumscribed rectangle is obtained for the clustered point cluster targets， which enables distinguishing between different types of targets such as cars and fences by combining the scattering intensity of the target. Finally， the validity of the algorithm proposed in this paper is verified by the test data in the field （a typical parking lot scene）.