Number of points of each scan in nuScenes

#1

I am using nuScenes-lidarseg dataset in my project, and I counted the number of points of each lidar scan file (34149 files of lidarseg trainval subset).

The number of points are just in these several values: [34272 34304 34336 34368 34400 34432 34464 34496 34528 34560 34592 34624 34656 34688 34720 34752 34784 34816 34848 34880]

Acorrding to the email from Holger, The nominal horizontal resolution is 1080 points (or 0.3333 deg) per ring, but the LiDAR can rotate slower or faster, so this number varies from 1071 to 1090, and the number of points of each scan file is [34272 : 32 : 34880]

For lidar scan files this number should vary in a range, because some laser rays may be lost. However, in nuScenes, all rays return and no point is dropped. Can anyone give some details on it? Are all the points in scan files trustable?

#2

Hi. The nominal horizontal resolution is 1080 points (or 0.3333 deg) per ring, but the LiDAR can rotate slower or faster, so this number varies from 1071 to 1090 as you pointed out. It is a multiple of 32 as all beams rotate at the same speed.

#3

Hi, @holger-motional,
Thanks for the reply.

When the horizontal resolution is 1080 points, the total number of laser rays should be 34560(=1080*32). But why all the rays can reture and all points are detected. For example, if the ray goes to the sky, no points should be detected. Or when there is rain or fog, more points should be lost. In this situation, the number of points may be about 30000 or less. However, in the data of nuScenes, there are 34560 points. No ray is dropped. Can you give some details on it?

Regards

#4

Yes, I believe the lidar sets all returned distances that are invalid to a value close to 0. That’s why it is important to filter points close to the sensor, which may often be the majority of the points.