Loop Closure not detected

classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|

Loop Closure not detected

rishavgh
Hello,

I am using RTABmap for visual odometry and mapping in a Gazebo world. The odometry is working well but the correct loop closure is not being detected.

In this database, the nodes with ID 1 and 1679 are quite similar and were close in the gazebo simulation but rtabmap did not detect a loop closure.

Here is the launch file. Please let me know what parameters need to be corrected. I have tried with RGBD/ProximityBySpace=true and RGBD/NeighborLinkRefining=true but it still did not detect a loop closure.
Reply | Threaded
Open this post in threaded view
|

Re: Loop Closure not detected

TOM_RB
This post was updated on .
CONTENTS DELETED
The author has deleted this message.
TOM SMITH
Reply | Threaded
Open this post in threaded view
|

Re: Loop Closure not detected

rishavgh
Thank you for your prompt reply, but I cannot see the lines you are asking to add.
Reply | Threaded
Open this post in threaded view
|

Re: Loop Closure not detected

TOM_RB
try add these parameters :

      param  name="Mem/IncrementalMemory" type="string" value="true"
      param name="Mem/InitWMWithAllNodes" type="string" value="true"

sorry for the previous empty massages
TOM SMITH
Reply | Threaded
Open this post in threaded view
|

Re: Loop Closure not detected

TOM_RB
In reply to this post by rishavgh
param name="Vis/CorGuessWinSize"       type="string" value="0"  
TOM SMITH
Reply | Threaded
Open this post in threaded view
|

Re: Loop Closure not detected

matlabbe
Administrator
You can also try with RGBD/LoopClosureReextractFeatures = true.

The problem is that mostly all features extracted for loop closure detection are outside the depth range:


With Mem/DepthAsMask=false and Kp/RoiRatios="0 0 0 0.45", this is fine for loop closure detection in this kind of environment, but not so for transformation estimation. Setting RGBD/LoopClosureReextractFeatures makes rtabmap reextract features with Vis/ parameters, thus using close features and forcing to extract only in depth range: