The number of 4D radar point clouds is large, and the bounding box of the point cloud can be obtained by clustering and calculating the orientation by referring to the processing method of the lidar.
Firstly, through DBSCAN clustering, the clustering result of the target is obtained. The red in the figure below is the point cloud that crosses the target directly in front, showing the use of a bird's-eye view.
In order to calculate the OBB of the clustered point cloud, the orientation of the target needs to be calculated first, and the PCA algorithm can be considered. Since the target is on a flat ground in most cases, it can be considered to use the ground projection point calculation (that is, only calculate x, y).
After obtaining the orientation vector vector of the clustered point cloud, rotate the x coordinate axis to be consistent with the vector. See references for methods.
The maximum and minimum coordinates of the point cloud after coordinate transformation represent the scale of the clustering target, that is,
length = maxL - minL
width = maxW - minW
You can also use the coordinates of the center point
centerX = (maxX + minL) / 2
centerY = (maxY + minY) / 2
The coordinates of the four corners can be calculated by the center point and the length and width, and then reversely rotated to obtain the center point and four corner points of the radar point cloud in the original coordinate system, and the bounding box of the point cloud can be obtained by connecting the four corner points.
Synchronization can obtain the length, width, orientation and other information of the clustered target.
References: