Re: RGBD SLAM and then RGB Localization

Posted by Kaifei on
URL: http://official-rtab-map-forum.206.s1.nabble.com/RGBD-SLAM-and-then-RGB-Localization-tp489p520.html

Hi,

    I ended up doing these steps to localize the RGB image:
      1. use memory_->computeLikelihood() to find the most similar signature (say it is signature N) in memory to the new RGB image (say it is signature M)
      2. Compute the transform between M and N using memory_->computeVisualTransform().
     (This is basically the loop closure codes that you have.)

    The step 1 works just fine, I can find a very similar signature in memory.
    However, when I do the second step, I need to do bowEpipolarGeometry internally in memory.
    The problem is, even though I have two very similar images (M and N), I always cannot get enough inliers.
    One example warning message is "Memory.cpp:1981::computeVisualTransform() Not enough inliers 9 < 20 ".
    I have changed the util3d codes to use different K for each image (because they are taken from different cameras).

     Do you have some suggestions on how to debug the EpipolarGeometry codes?
     Thank you very much!



Thanks!