recover images time stamp problem

[LauZanMo]

Hello, I'm using e2calib for imu-ecam chain calibration and I recover images from event camera successfully. Kalibr calculates the intrinsic of my event camera correctly(reprojection error RMS 0.15pix) but when it comes to extrinsic, the calibration can not converge with huge reprojection error and wrong extrinsic results. The event camera is synchronized with IMU using hardware trigger. I check the synchronization for a few times but every thing goes well. So I wonder if I miss some details about your project in taking stamps for events and hope you can help me. THX

[magehrig]

Hello @LauZanMo,

It's important to note that a low reprojection error doesn't necessarily indicate accurate calibration. To ensure accuracy, it's essential to have comprehensive coverage of detected corners within the camera frame, as this makes it more challenging to achieve a low reprojection error.

In our experience, we encountered similar issues when attempting IMU-eCam calibration with a low-cost IMU, specifically one integrated within the camera case. We haven't tested this process with a higher-quality IMU. Have you had the chance to try it with a better IMU?

[LauZanMo]

Prophesee EVK4 HD(event camera) and ADIS16465(IMU) is mounted on my test suite. However, I use kalibr intrinsic result to run pure vo (ORB3) in the e2calib recover images and the trajectory result is bad with only pure rotation, as I only do pure translation during the calibration. Maybe the intrinsic of output is also wrong although the reprojection error is low. I think the problem may comes from the wrong stamps of events. Do you have any idea?

[magehrig]

When you look at the report from Kalibr, do you see good coverage of the detected pattern? That is, has the pattern been detected throughout the whole field of view?

See: https://github.com/ethz-asl/kalibr/wiki/multiple-camera-calibration#3-the-output

Follow the instructions in this video: https://github.com/ethz-asl/kalibr/wiki/camera-imu-calibration