How do you approach the calibration of multiple sensors in an autonomous vehicle to ensure accurate data integration?
Instruction: Describe the process and challenges of sensor calibration.
Updated
Example Answer
I would separate sensor configuration from calibration. Camera intrinsics describe projection and lens distortion; extrinsics describe transforms between sensors and vehicle coordinates. Time offsets also matter when a moving scene is observed by sensors with different timestamps. I would estimate the required parameters with suitable targets or other validated methods, check observability and residuals, and verify on separate data. Changing exposure is not the same as calibrating camera-to-LiDAR geometry.
Make it your own
Illustrative calibration record: frame names, transform direction, units, camera matrix, distortion model, timestamp convention and held-out reprojection/alignment errors.
Why this works
A reversed transform or unnoticed time offset can create plausible but misplaced objects. Follow-up: why might one stationary target pose be insufficient to estimate every parameter?
Sources: OpenCV calibration and reconstruction, OpenCV camera calibration.