A practical deflicker approach is to estimate each frame’s overall luminance, smooth those values over time, then adjust each image to match the smoothed curve.

One community-suggested method is:

1. Measure frame brightness, such as by summing or averaging pixel luminance.
2. Compare adjacent frames to estimate brightness change.
3. Ignore large per-pixel differences when doing this so subject motion doesn’t skew the result.
4. Build a brightness curve for the whole sequence.
5. Apply a moving average (or similar temporal smoothing) to that curve.
6. Adjust each frame’s exposure/brightness to follow the smoothed result.

This works because true flicker is usually short-term variation, while sunrise/sunset exposure change is gradual and should be preserved.

If you want an existing tool rather than writing your own, one answer points to a Python project made for timelapse processing and deflickering: timelapsepy. It can also interpolate missing frames.

In short: use per-frame luminance estimation plus temporal smoothing, while rejecting motion outliers.