How can I convert 3D bounding boxes to 2D bounding boxes on the camera?

#1

Hello! How can I convert 3D bounding boxes in Lidar coordinate to 2D bounding boxes on the camera? When I determine which cameras the box is in, I have used box_in_image() function. But some boxes are still can’t find accurate camera positions. Do you have accurate relative position relation between cameras coordinate and lidar coordinate or cameras coordinate and lidar coordinate?

1 Like
#2

Hi @nmll, you could take inspiration from our method which maps a pointcloud to the image plane: https://github.com/nutonomy/nuscenes-devkit/blob/84ee1f3ad767a0ac5953fdd2f2d622bef46bedb0/python-sdk/nuscenes/nuscenes.py#L834

1 Like
#3

OK, Thanks very much !

1 Like
#4

Hi, I tried that link to project 3D bounding boxes in Lidar Coordinate to 2D bouding boxes on the camera. however, It doesn’t work. But when I tried that link to project lidar point cloud onto images, it works. So I guess 3D bounding boxes and LiDAR point clouds are not in the same coordinate…
is there any other solution??

#5

@Konyul_Park pls refer to this: https://github.com/nutonomy/nuscenes-devkit/issues/793#issuecomment-1194881752

1 Like