Carla: add semantic label to lidar point cloud

[Tom-Huang]

The projection of semantic label on lidar data is not accurate

I do the following to add semantic labels to point cloud data from lidar sensor in carla:
1.I get the intrinsics as stated in #56.
2.Project 3d points into semantic segmentation image of different views(front, left, right, rear).

effect:
Totally, the road, buildings and surrounding look good, but the vehicles and pedestrains seem to have some errors.

The general effect:

The local accuracy is not accurate:

The corresponding semantic images of front, left, right and rear views:
front:

left:

right:

rear:

The problem is: the lidar data points cast onto the vehicle does not have the same shape as the vehicle. As a result, the projection has deviation. Does anybody have any idea of how to improve the accuracy?

[user025]

#1010 might be helpful.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[heshameraqi]

Could you please share your code for projection? It will be very helpful.

[Tom-Huang]

Could you please share your code for projection? It will be very helpful.

You can check my github repo carla_data_collection. Hope it will help!

[marcgpuig]

This feature has been added in CARLA in #3131, with better performance and more accurate bounces in vehicles and pedestrians.