intensity calibration

[yuhui-xie]

Hi, it's good to see your open project. May I ask how do you deal with the intensity calibration?

[yorklyb]

Hi. For the Livox MID-40 LiDAR, the intensity is directly available. Please refer to the paper of livox_loam(https://arxiv.org/pdf/1909.06700.pdf). The Velodyne 16 and 32 LiDAR return the 256 calibrated reflectivity which is independent of laser power and distance. (https://velodynelidar.com/wp-content/uploads/2019/12/63-9243-Rev-E-VLP-16-User-Manual.pdf) (https://icave2.cse.buffalo.edu/resources/sensor-modeling/VLP32CManual.pdf)

I feel like you probably already knew these given that you asked such a professional question. I introduced the above contents before answering your question since I hope that the other users interested in this question could also see it.

We left a customizable threshold as the 'intensity' value returned by different LiDAR models could differ. That is, sometimes the calibration is needed, sometimes it doesn't. It depends on the LiDAR model. When developing the framework, we don't want our system to be coupled with any specific LiDAR model.

Moreover, in the pipelines of our framework, we are actually only concerned with the contrast since each intensity value is transferred into a specific color through colormap (https://arxiv.org/pdf/1509.03700.pdf). And the intensity image is transferred into a grayscale image then. So it doesn't really matter if the calibrated reflectivity is adopted or the intensity is adopted. As long as the contrast between the black and white regions is obvious, the quality of the grayscale image will be good.

I admit that the algorithm will be more rigorous if we always utilize the calibrated reflectivity. However, take the Livox MID-40 LiDAR as an example, we have to add a new function to calibrate the intensity since it only returns intensity, which will make the system slower. Also, it is not a necessary process in the current implementation considering that a threshold already solves everything.

[yuhui-xie]

Hi. For the Livox MID-40 LiDAR, the intensity is directly available. Please refer to the paper of livox_loam(https://arxiv.org/pdf/1909.06700.pdf). The Velodyne 16 and 32 LiDAR return the 256 calibrated reflectivity which is independent of laser power and distance. (https://velodynelidar.com/wp-content/uploads/2019/12/63-9243-Rev-E-VLP-16-User-Manual.pdf) (https://icave2.cse.buffalo.edu/resources/sensor-modeling/VLP32CManual.pdf)

I feel like you probably already knew these given that you asked such a professional question. I introduced the above contents before answering your question since I hope that the other users interested in this question could also see it.

We left a customizable threshold as the 'intensity' value returned by different LiDAR models could differ. That is, sometimes the calibration is needed, sometimes it doesn't. It depends on the LiDAR model. When developing the framework, we don't want our system to be coupled with any specific LiDAR model.

Moreover, in the pipelines of our framework, we are actually only concerned with the contrast since each intensity value is transferred into a specific color through colormap (https://arxiv.org/pdf/1509.03700.pdf). And the intensity image is transferred into a grayscale image then. So it doesn't really matter if the calibrated reflectivity is adopted or the intensity is adopted. As long as the contrast between the black and white regions is obvious, the quality of the grayscale image will be good.

I admit that the algorithm will be more rigorous if we always utilize the calibrated reflectivity. However, take the Livox MID-40 LiDAR as an example, we have to add a new function to calibrate the intensity since it only returns intensity, which will make the system slower. Also, it is not a necessary process in the current implementation considering that a threshold already solves everything.

Thank your for your comment！ Well actually I only knew there exist a lidar intensity model which is "intensity \propto cos(Incident angle) / (range)^2"，and I found that a open source work has mentioned it in paper but drop it in code, which really confuses me. Now I know the calibration has been done already for Livox and VlP, thank u very much!!