Image motion compensation driven by optical flow

Abstract

The image motion compensation prototype tracks the scene by translating the sensor on a piezo stage, with a profile computed from the orbital parameters. The profile, however, is a prediction, and any mismatch between predicted and actual motion leaves residual smear.

The thesis closes the loop on measurement: it estimates the apparent motion from the images themselves, with optical flow on low exposure pairs, and uses that estimate to drive the stage.

Objectives

1. Estimate the apparent motion from low exposure image pairs, in time for the command.
2. Close the stage control loop on that estimate.
3. Compare the residual smear with the open loop, predicted profile case.
4. Identify the limits of the estimate when the scene has little structure.

Methodology

1. Selection and tuning of an optical flow method with a cost compatible with the command rate.
2. Characterisation of the overall latency, from acquisition to stage motion.
3. Control synthesis that accounts for that latency.
4. Bench tests with the motorised slide, on scenes with different spatial frequency content.
5. Residual smear measured with the modulation transfer function, comparing open and closed loop.

Expected contributions

1. A working closed loop compensation chain on the bench.
2. A quantitative comparison between open and closed loop on the same scene.
3. Characterisation of the scenes where the estimate is unreliable, and the fallback behaviour.

Tools

Closed loop biaxial piezo stage, bench camera and motorised slide, Python and OpenCV.