Preprocessing

Preprocessing the FAAM videos

The Aircraft videos, unfortunately, appear particularly noisy for the front-facing camera. This noise would affect any auto cloud edge detection we might apply. The script processvids.py searches in a faam-video folder for mp4 files matching the date and camera given and produces a clean, timestamped video.

nosiy video

Initial work revealed the camera metadata time was offset from the aircraft instrument data by around 10 seconds, manually for each day. This was estimated by plotting the aircraft roll data. We created a csv of lag times (input_data/lag_times.csv) for each day, and this is used to correct the timestamp.

roll video

We take each frame and apply a bilateral filter across all three channels RGB to create a cleaned version of the video with less noise. A more aggressive filter could be applied, but then cloud detail is lost.

clean video

A standalone version of this tool can be used with additional functionality to clip and speed up the clips.