./rtabmap-console with SIFT or ORB

[m.omar82]

dear friends
how can use ./rtabmap-console with SIFT or ORB

thanks

[matlabbe]

SIFT:
$ ./rtabmap-console -Kp/DetectorStrategy 1 -Kp/NNStrategy 1 -Kp/NndrRatio 0.8
ORB:
$ ./rtabmap-console -Kp/DetectorStrategy 2 -Kp/NNStrategy 3 -Kp/NndrRatio 0.9

• Kp/DetectorStrategy : "0=SURF 1=SIFT 2=ORB 3=FAST/FREAK 4=FAST/BRIEF 5=GFTT/FREAK 6=GFTT/BRIEF 7=BRISK."
• Kp/NNStrategy : "kNNFlannNaive=0, kNNFlannKdTree=1, kNNFlannLSH=2, kNNBruteForce=3, kNNBruteForceGPU=4"
• Kp/NndrRatio : "NNDR ratio (A matching pair is detected, if its distance is closer than X times the distance of the second nearest neighbor.)"

Parameters that can be set:

$ ./rtabmap-console -default_params

Description of the parameters can be found in Parameters.h. Some examples on the loop closure detection benchmark page.

[m.omar82]

 Thanks your are the best

[m.omar82]

Dear matlabbe
How did you chose the optimal parameters for SURF?  if we use the default parameters will get recall 79%

[matlabbe]

I use the default parameters for SURF, except the hessian/response threshold that I change accordingly to the image size. See Sections III-A and IV paragraph 2 of this paper. The hessian threshold should be low enough to have around 400 features extracted from the images and high enough to avoid bad features or too many features extracted (with images at high resolution). The 400 maximum features parameter was fixed experimentally, mostly with the challenging UdeS dataset. This parameter can be lowered to increase the WM size manageable in real-time.

[m.omar82]

Big thanks, and Happy New year