I need rotation and translation matrix values to align RADAR_FRONT data to the CAM_FRONT images.
I searched the devkit code but did not find these two matrices.
take a look at the four steps in https://github.com/nutonomy/nuscenes-devkit/blob/5325d1b400950f777cd701bdd5e30a9d57d2eaa8/python-sdk/nuscenes/nuscenes.py#L784 . We don’t work with matrices, but rather with a quaternion (rotation) and a translation vector.
I am working with the radar data. Radar data can be converted to Camera frame by rigid body transformation matrix btw Radar and Camera, i.e., [R|t] through the vehicle coordinate system. However, the height dimension of the Radar data (Z axis for the RADAR_FRONT) is not accurate as indicated in some papers. Therefore, if your aim is to project radar points into the image frame using the camera intrinsic matrix, you can see wrong radar to pixel associations.
Hi, the radar is mounted horizontally and the z value is simply that of the sensor. So if there are minor calibration errors, the values will be off.
I was talking about the x,y,z coordinates of the radar points. The z value is the height of a radar point and that value is not used in most papers since they indicate that the z values are not accurate. Sorry but I could not understand your comment. Could you please clarify it?
Yes. z is not just not accurate, it is basically not provided. We just pad it with a 0, which is equivalent to the radar sensor height after moving it to global coordinates.