Inconsistent motion detection

[ZSasha]

Hi guys,
Could you please help me with my situation - it just drives me crazy?

While using the "Testing & Tuning" tool to troubleshoot missing motion events, I came across a very strange situation.
The same video footage triggered motion events twice consecutively, but the third time, it completely missed the event (the square around the white car remained light-yellow).

Why is this happening? It's the same video replayed multiple times, and I'm testing simple motion detection, not AI.

I thought it was straightforward mathematics, with a certain number of pixel changes, so the results should be consistent.
However, this doesn't seem to be the case. With the lowest thresholds set (visible in the settings), how can I ensure it always works?

Additionally, I observed that playing the video in reverse almost instantly triggers the motion event (as it should according to my settings), but playing it forward results in this issue.

Here is the link to the video - https://youtu.be/svqD4N5PGSo

My camera settings:

settings2.jpg (33.29 KiB) Viewed 2243 times

settings1.jpg (52.11 KiB) Viewed 2243 times

Is it a bug or I am doing something wrong?

Your help would be greatly appreciated!

[TimG]

Interesting. I didn't expect that ! Can you ask Support to explain it ?

[Pogo]

If you are primarily concerned with refining Motion detection you will be much better off disabling Object detection or eliminating it altogether unless it becomes a special requirement as a 'poor man's AI' solution to minimizing nuisance alerts in certain situations. (The size and contrast settings still apply to basic Motion detection.)

I would also change the 'Method' to Simple for basic Motion detection.

The Edge Vector detection method may be the source of the back-assward triggering issue when playing a video in reverse. 'Simple' behaves a bit differently and is usually what I use for motion detection. Edge Vector has been noted as better suited to Object Detection and AI setups. It supposedly responds to the leading edge of anything it sees vs. relying on pixel numbers or sizes..., in other words, objects vs. general motion.

In the case of troubleshooting and tuning with reverse video using Edge Vector it would only seem reasonable that it would behave a bit differently triggering on a trailing vs. leading edge of a frame. Whether it does or doesn't seems beside the point. Is there an obvious purpose for the reverse analysis?

Edited to add: The .5 Make time once every 30 seconds pretty much equals the triggers shown in the video with the exception of perhaps the last one? The 'Break' time may also need adjustment for more appropriate triggering behavior.