What supervised classification algorithms can be suitable for detecting reflective surfaces in 3D Laser Scan point cloud?

3 posts / 0 new
Last post
JohnDuke
Offline
Last seen: 11 months 1 week ago
Joined: 12.08.2017 - 16:29
What supervised classification algorithms can be suitable for detecting reflective surfaces in 3D Laser Scan point cloud?

Hello folks,

I am doing a project that will involve detection of reflective surfaces in 3D laser scan point cloud. What kind of supervised classification algorithms do you think will be suitable? SVM, decision tree, KNN, ANN, logistic regression, etc. I am thinking of using SVM, ANN, and LR.

I have PCL and OpenCV installed on my machine.

Thanks

Forum Schlüsselworte: 
rsteffen
Offline
Last seen: 11 months 1 week ago
Joined: 17.02.2009 - 11:00
Detection of reflective surfaces

Dear John,

the classification of different surfaces depending on the reflectivity value needs a normalization of the reflectivity due to the surface angle first. For terrestrial scans this can be done by estimating the surface normals first and assume a reflectivity model like lambertian and normalize the reflectivity.
To get good results for the classifier (applying a supervided classifier like SVM or logistic regression) you need a model for spatial relation ship between the points. Actually, I do not know a good implementation/model to do that.

Best regards,
Richard

JohnDuke
Offline
Last seen: 11 months 1 week ago
Joined: 12.08.2017 - 16:29
Dear Richard, Many thanks for

Dear Richard,

Many thanks for your contribution! I will search if I can find a model for the spatial relationship between the points. I was also thinking of using a training data of photographs to train the classifier. So that reflective surfaces can be detected from the photograph of same scene automatically, then look for a way to relate the position in the photograph to the position in the terrestrial laser scan.

Best regards
John