Skip to content

Too many false positives #19

@5559175

Description

@5559175

I have been using this more and it is working quite well to detect actual meteors!

I have quite a nice automated setup to capture a stream in a better way to avoid duplicate timestamps, long frames and general strange FPS readings. I will keep the description of this and the other automation I have wrapped around it out of this report. Instead I have added it as a discussion/show and tell.

Anyway, I am now seeing a good deal of false-positives being detected.

There is either nothing at all, or they are definite aircraft on a good majority of clips produced.

Aircraft are quite obvious to my eye as they will almost always have a periodical/regular/predictable flash of a tail/wing-light. I don't think the model/algorithm must be taking this sort of thing into account?

Most meteors don't seem to flash quite so regularly/predictably as an aircraft strobe light when breaking up so I don't know if this is perhaps something that can be added in?

In these examples I was using ClipToolkit.py to create clips from the detections.json with "--enable-filter-rules" and here are the results (to my eye at least):

34 detections, only 14 correct
20 detections, only 5 correct
15 detections, none correct

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions