I'm scanning video that has time codes burned in at capture by ffmpeg. A few times I have noticed thumbnails that depict motion that happened long before the motion event.
For example, event 278 started at 17:44:20, and ended at 17:44:25. Thumbnail 278 is a frame from 17:31:16. (Event 277 started at 17:25:08, and ended at 17:25:19.)
These errant thumbnails tend to show birds or bees flying through the frame. Since my camera is mounted up high, things that fly can get a lot closer to the camera than things that can't fly. So a bird appears big and fast, just by virtue of being closer to the camera, and more in the foreground, and I guess such a frame gets a very high score. But for whatever reason, the model ultimately decides that the bird was a non-event, and continues on.
But then, if the next bona fide motion event does not contain any frames with a score higher than the bird, self._highframe will still be pointing to the bird frame when the next thumbnail is saved.
I'm scanning video that has time codes burned in at capture by ffmpeg. A few times I have noticed thumbnails that depict motion that happened long before the motion event.
For example, event 278 started at 17:44:20, and ended at 17:44:25. Thumbnail 278 is a frame from 17:31:16. (Event 277 started at 17:25:08, and ended at 17:25:19.)
These errant thumbnails tend to show birds or bees flying through the frame. Since my camera is mounted up high, things that fly can get a lot closer to the camera than things that can't fly. So a bird appears big and fast, just by virtue of being closer to the camera, and more in the foreground, and I guess such a frame gets a very high score. But for whatever reason, the model ultimately decides that the bird was a non-event, and continues on.
But then, if the next bona fide motion event does not contain any frames with a score higher than the bird,
self._highframewill still be pointing to the bird frame when the next thumbnail is saved.