ollie-nye
commented
4 years ago On startup, each camera now drops the first 60 frames to allow for environment adjustment, then captures 30 frames and averages them. This forms an initial model, which is combined with a running average and used against the overall comparison for each frame
Current background is a moving average. It stops marking objects as interesting after they've been in the same place, bit it is easily conceivable that someone might stand in the same place for a while.
Investigate taking a set of environment screenshots, and combining them with a rolling average. Possibly incorporate general brightness of both frames, and frame patterns/trends.