Optimization failed in kalibr_calibration_imu_camera about kalibr HOT 10 CLOSED

Please increase the timeOffsetPadding (https://github.com/ethz-asl/kalibr/blob/master/aslam_offline_calibration/kalibr/python/kalibr_calibrate_imu_camera#L171) and see whether this fixes your issue. In general, this parameter is the margin within the temporal offset is allowed to vary.

from kalibr.

PS: It is possible to use the --timeoffset-padding argument when running kalibr_calibrate_imu_camera.

from kalibr.

Please increase the timeOffsetPadding (https://github.com/ethz-asl/kalibr/blob/master/aslam_offline_calibration/kalibr/python/kalibr_calibrate_imu_camera#L171) and see whether this fixes your issue. In general, this parameter is the margin within the temporal offset is allowed to vary.

thanks, solved it with changing to 100ms

from kalibr.

Yes, it fixes. Is there any rule of setting this parameter, or just increase it when the same issue occurs.

from kalibr.

It is a trade-off between computation time and robustness: If you set the margins conservatively large, it will result in increased computation time, if you set it too low, the estimate might travel across knot boundaries in an unpredicted manner, violating the precomputed sparsity pattern. Please see the implementation section in Paul's paper for details on that (https://3234f89137bccf2ede29cc86e315c75116020d70.googledrive.com/host/0B64GJ60h3Ai1MVVwWTZwekhtcFU/publications/bib/furgale_iros13.pdf).
The "trajectory" of your estimate of the temporal offset during optimization will depend on the shape of the cost surface which in turn is related to the very motion you performed. Hence, it is difficult to provide a rule of thumb here.

from kalibr.

Got it, thank you Joern

from kalibr.

@francescoli Out of curiosity (actually facing this problem myself) what value did you increase the temporal offset to?

from kalibr.

While this parameter might need some adaptation, repeated problems setting the margins correctly often hint to issues in the data, for example due to spurious measurements, incorrect scaling or incorrect timestamping.
Please consider running the calibration with the --verbose option (https://github.com/ethz-asl/kalibr/blob/master/aslam_offline_calibration/kalibr/python/kalibr_calibrate_imu_camera#L92), which will plot the cross correlation used for finding an initial estimate (https://github.com/ethz-asl/kalibr/blob/master/aslam_offline_calibration/kalibr/python/kalibr_imu_camera_calibration/IccSensors.py#L217-L229). If the plot does not exhibit a distinct peak, please specifically check your gyroscope data, since this is being used here.

from kalibr.

Hello,rehderj , I also met this problem, I want to know the value timeOffsetPadding should be what value?
maybe 100?1000? or whatever?

from kalibr.

@gethubwy I've been using a value of 0.1 (100 ms), as per this issue. Note, however, that this is simply a buffer margin for the timestamps. Oftentimes, it might be helpful to review your dataset and ensure that there are no problems there instead of changing this value.

from kalibr.