Hi,
Here is what I have encountered when I run the code which you provided "Project 3D Velodyne points to 2D camera Image ", I got error message.
I put the code you provided in ipython notebook inside one file called, "projectVelo.py", and I only changed the path to velodyne and image file.

Here is the error message:

Traceback (most recent call last):
File "projectVelo.py", line 60, in
img, pnt, c_ = res.velo_projection_frame(v_fov=v_fov, h_fov=h_fov)
File "/home/keisha_tsai/Downloads/KITTI_Tutorial-master/kitti_foundation.py", line 611, in velo_projection_frame
res, c_ = self.__velo_2_img_projection(velo_gen)
File "/home/keisha_tsai/Downloads/KITTI_Tutorial-master/kitti_foundation.py", line 486, in _velo_2_img_projection
xyz_v, c = self.__point_matrix(points)
File "/home/keisha_tsai/Downloads/KITTI_Tutorial-master/kitti_foundation.py", line 305, in __point_matrix
self.__points_filter(points)
File "/home/keisha_tsai/Downloads/KITTI_Tutorial-master/kitti_foundation.py", line 368, in __points_filter
self.__upload_points(points)
File "/home/keisha_tsai/Downloads/KITTI_Tutorial-master/kitti_foundation.py", line 296, in __upload_points
self.__x = points[:, 0]
TypeError: 'generator' object has no attribute 'getitem'

Hope anyone can help me to solve this problem.
Thank you.

I found something illogical in function velo_points_2_pano of Convert_Velo_2_Pano_detail.ipynb;

def velo_points_2_pano(points, v_res, h_res, v_fov, h_fov, depth=False):
    # Projecting to 2D
    x = points[:, 0]
    y = points[:, 1]
    z = points[:, 2]
    dist = np.sqrt(x ** 2 + y ** 2 + z ** 2)

    # project point cloud to 2D point map
    y_img = -(np.arctan2(z, dist) / (v_res * (np.pi / 180)))

Should dist = np.sqrt(x ** 2 + y ** 2) rather than dist = np.sqrt(x ** 2 + y ** 2 + z ** 2)? I discussed with ChatGPT4,0 who also agrees with me.

What is the difference between "ax, ay" and "af, al" in the oxts about IMU?

Hi, thanks for your sharing code first.
However, when I try your code to realize the velodyne cloud points projection to image, I found there are mismatch between the cloud points and the corresponding image object, such as the cyclist of "2011_09_26_drive_0005_sync" dataset.

You can see the points cloud of the cyclist go through the side car.
I don't know why, maybe the original calibration matrix is not correct? Or the calculation is not right, because I am not understand the calibration matrix very well in "calib_cam_to_cam.txt".
Thanks!

Hi Great work!
I was using Convert_Velo_2_Pano.ipynb for my Indian Driving Data(IDD) and the lidar used for this dataset is Velodyne VLP-16.I changed the angular resolution and hfov and vfov accordingly,but I got a very bad result.

But when I use the code on Kitti dataset for Velodyne HDL-64E (with the respective resoluion and fov)
I got much better results.

Can anyone suggest what is going wrong?
Is the conversion also dependent on any other factors based on lidar?
Thank you

Hi,
Thanks for the tutorial.
Here is what I have encountered when I run the code which you provided "Project 3D Velodyne points to 2D camera Image ", I got error message.
I put the code you provided in ipython notebook inside one file called, "projectVelo.py", and I only changed the path to velodyne and image file.

Traceback (most recent call last):
File "projectVelo.py", line 54, in
res = Kitti_util(frame=128, camera_path=image_path, velo_path=velo_path, v2c_path=v2c_filepath, c2c_path=c2c_filepath)
File "/home/keisha_tsai/Downloads/KITTI_Tutorial-master/kitti_foundation.py", line 235, in init
super().init(frame, velo_path, camera_path, img_type, v2c_path, c2c_path, xml_path)
TypeError: super() takes at least 1 argument (0 given)

Could you help me to clarify the error? Thank you.

I am getting an error while trying to execute the display_groundtruth.ipynb notebook.
In cell number 4

please it will be highly appreciated if you can give me a clue about it ?

Hey there,

There's an issue with your ground truth script. It links back to the parsing script, itertools.izip does not work. I changed it to zip (python2 -> python3) but then the main loop fails at:

for i, j in zip(tracklet2d, type_[frame]):
    for k in line_order:    
        cv2.line(image, (int(i[0][k[0]]), int(i[1][k[0]])), (int(i[0][k[1]]), int(i[1][k[1]])), type_c[j], 2)

In the projection part, your algorithm is to remove the points outside the camera frame by using distances and angles. But I cannot find the theory which support this way.

Actually we can use the whole pointcloud with Extrinsic and Intrinsic matrix to get the results of [x, y, w], then remove the points with w<0. This way is more understandable.

There are 9 values for the rotation matrix in calib_velo_to_cam.txt of KITTI dataset. Can anyone interpret these 9 values??

Could you please provide the calibration file?
Thank you!

We should pay attention that the Elevation angle of Velodyne HDL64 is not even, the upper 32 beams are from 2.0_deg to -8.5_deg, and the lower 32 beams are from -8.87_deg to -24.87_deg.