I am working on lane line detection using lidar point clouds and using sliding window to detect lane lines. As lane lines have higher intensity values compared to asphalt, we can use the intensity values to differentiate lane lines from the low intensity non-lane-line points. However my lane detection suffers from noisy points i.e. high intensity non-lane line points. I've tried intensity thresholding, and statistical outlier removal based on intensity, but they don't seem to work as I am dealing with some pretty noisy point clouds. Please suggest some non-AI based methods which i can use to get rid of the noisy points.
Good morning,
there are many options for decimating lidar point cloud.
I am of the simpler tools is the SOR filter in the open-source software "CloudCompare". Here is the link to the instructions : https://www.cloudcompare.org/doc/wiki/index.php/SOR_filter
Maybe this paper will be usefull for you:
Article An Efficient Adaptive Noise Removal Filter on Range Images f...
Kind regards,
Jakub
I agree with Jakub Miszczyszyn and wanted to add you could also use software Rapidlasso LAStools to remove noise. If you are using your own home grown code use a spherical filter that counts points and thresholds of the number of points within a set radius with a sphere for every point. You could add a control for intensity also but that would be up to you.
Good question, I have no exact answer, but making use of the linear feature (adding geometry) that the line forms should make your estimation more robust. I have seen Hough transform being used in tracing linear pieces on the ground (deined by the 3D geometry).
- Badges
- Science topic