BI + CP AI Motion tuning trouble

[bug99]

I am trying to figure out / tune two of my front cameras with regards to motion triggers using AI. Most of the settings are under each camera Trigger/motions sensor/configure. I think this issue is common, applying to many people’s BI-AI setups.

There are a few things that I haven’t been able to make sense out of. Let me try to describe the main issue (with some images).

If a person walks from my front door between my two cars, out of the frame, waits a few seconds and then returns (like 10 seconds later), I only get the return trip as a stored alert clip (image 2 below). When I analyze both the leaving and the return video with AI, I see similar ~80%+ orange framed triggered “person”. Why is the first one not a proper trigger to alert clip, but the second is?

I have messed around a little bit with number of frames (2-6) and time (750ms to 333ms), and the contrast sensitivity, but so far I have seen little effect. I thought the issue was related to Hikvision, but the Dahua has a similar issue (with similar settings). [note: I discovered that the Dahua (second set) does not actually capture a person clip at all (neither going or leaving). The clip only captures and tags a car on the road. The person was collateral].

Fundamental questions:

• Is the primary motion trigger coming from BI, and then being sent to AI? Is this why we configure min obj size, min contrast, min duration, etc here vs leaving it the AI to decide? That then sounds like a balance. If so, I think we need some guidance on tradeoffs, including with regards to “Advanced/object detection” settings.
• Notice the below images. The first has a min obj size, but no max. However, when Advanced/Object detection is selected, a maximum (or at least a reduced border) appears (see image window with rectangular doughnut). I think the border of the doughnut is the 67% referenced in the third screen shot.
• Should “Advanced/object detection” normally be used with AI? For what purpose besides zone creation and direction (A>B)?
• If the AI is set to analyze 4 images for 500ms, when does it start the first of the four? Is it the first frame from BI, from the above logic processing? Assuming 10fps, does the AI then ignore the next 5 frames (most likely better than the first one out of the gate) and then start on the 6th frame, etc? Is it looking at the main video stream of the sub-stream?
• Does “Include JPGs on all clips & timeline” do anything useful (under the record tab, attached image 5), or is it just filling up the Alerts tab?