Firstly, i'd like to thank for your great contribution. But i have some question about this software.
I install the rtab-map with ros: $sudo apt-get install ros-kinetic-rtabmap-ros
And then i use the command "$ rtabmap" to start the program, and I get the interface like this:
I had tried using the RGB-D datasets to generate the dense map, which is pretty good.
However, when i want to use the stereo images to generate the map, it warns that i need to calibrate like the following picture:
But actually, the stereo images i got is simulation data(have no distortion and don't need to do stereo calibration). so i wonder if it possible to run it directly without the process of calibration, or maybe i could manually provide a calibration parameters file(just like using RGB-D image)? If i could manually provide the calibration file, what kind of the format should it have?
And I wonder if it possible to reload the *.db and export the pointclouds through the terminal command?
Acutally, my computer's computing resource is not strong enough. So, it would crash when I export the pointclouds( more than one hundred million points). And I decide to use the server to do that with docker. But I guess I cannot use the GUI on the docker?
Thx for replying!
In order to increase the map's accuracy, I want to use the ground truth as the input of the odmetry (because I guess the more accurate the poses are ,the more accurate the map is). So I do the following setting:
I add the groundtruth as the input of the odmetry with the format in x,y,z,r,p,y.
After the test, I export the poses. But I find that the exported poses is not the same with the odmetry I input. I thought that as I have already provided the odmetry(poses), then, the whole process would carry on based on the odmetry I provide, and the exported poses should be the same with odmetry or at least their trajectories should be coincident.
Could you please answer my doubt? And is it feasible to increase the map's accuracy for providing the accurate odmetry?
If there are loop closures or proximity detections, the final poses won't be the same. If neighbor link refining is enabled, the poses won't be the same either. Just to debug that these poses are correctly recorded in the database, you can do Edit->Download Graph only, then choose "Not optimized". This should be the graph with your input poses.
I meet another problem. I install the docker of rtabmap, and use the command $rtabmap to get the interface and then I run the rtabmap. After finishing the process, I want to export the 3D pointcloud. I use the File->Export 3d clouds. And then I set the parameter voxel size to 0, then I export the pointclouds. However, I met the core dumped. I tried several times, and the result is the same. Actually, I met this problem several times, sometimes it's ok, but sometimes it would crash.
The log information is shown in the following picture:
I wonder why? Is it because the data is too large that the cpu cannot hold it?