Open hdevillepoix opened 4 years ago
Hadrien, do you mean the rolling buffer is too small? That is 15s now, set in config file.
I'm don't think the issue comes from the size of the rolling buffer. Having said that 15s seems short. Does this mean a >15s event is considered too long to be a fireball? (we have seen 5 fireballs longer than that so far.) I would increase to 30s it if we have enough memory for it.
Though I think the real problem is not there.
If we have observed a series of frames, with X a frame for which an event has been detected, and 0 a frame without detection:
00000XXXXXXXXXXXXXXXXX00000
.
Some of the 0s at the beginning and at the end obviously contain the fireball, but are too low SNR to be detected. They need to be saved for further analysis, but with a larger buffer than what is currently used.
Example: 000000000000XXXXXXXXXXXXXXXXX0000000000000
frames need to be saved instead of 00000XXXXXXXXXXXXXXXXX00000
.
Problem is likely around: https://github.com/desertfireballnetwork/freeture_DFN/blob/master/src/DetThread.cpp#L215
When an event is detected there is a bit of a buffer of extra frames saved to account for when the object is too faint to be detected. For some events this buffer is not big enough, raw frames containing the fireball are missed.