search

raulmur / ORB_SLAM2

Real-Time SLAM for Monocular, Stereo and RGB-D Cameras, with Loop Detection and Relocalization Capabilities
Other
9.5k stars 4.71k forks source link

Run stereo ORBSLAM2，why is the coordinates so small? #794

Open Wangfeng2394 opened 5 years ago

Wangfeng2394 commented 5 years ago

I run stereo orbslam2
And compare it with cartographer What I can confirm is that cartographer can be used as an accurate standard.

The trajectory of two methods above as follows： 11111

222222222222222

thohemp commented 5 years ago

Have you adapt the specific calibration information for your camera?

SiyuanHuang95 commented 5 years ago

Sorry for my dummy questions: But how could you get the trajectory? Any hint is wellcome! Did you reconstruct it from KeyFrameTrajectory.txt ?

thohemp commented 5 years ago

@AnfangRobkit Yes, check this: https://github.com/MichaelGrupp/evo

SiyuanHuang95 commented 5 years ago

@AnfangRobkit Yes, check this: https://github.com/MichaelGrupp/evo

Great hints! And thanks for your reply!

One more question, the KeyFrame has obvious fewer points than the ground truth, but due to the x-axis is timestamp, so it did not affect the result? Am I right?

Anyway, still great thx for your information.!

mathuse commented 4 years ago

@AnfangRobkit Yes, check this: https://github.com/MichaelGrupp/evo

I have the stereo camera, and I migrate the code to my code. I find some problem,the first problem is that what time the program calculate the "pKF->mFeatVec" when program first enter the "TrackReferenceKeyFrame/SearchByBoW". I slove it by adding "mCurrentFrame->ComputeBoW();" before "StereoInitialization/new KeyFrame" but it is not perfect, because I have successfully tested stereo camera using the executable file compiled from ORB-SLAM2 code. The second problem is "std::bad_alloc" when I run the my executing file, and I have find the location of this problem at "TrackReferenceKeyFrame/PoseOptimization/optimizer.optimize(its[it])". I want to know the reason.Can you help me? By the way, I have tried a lot of ways, but none of them worked.