AI (CodeProject) working hard but why?

Post Reply
Ferguson
Posts: 33
Joined: Sat Feb 22, 2020 1:32 am

AI (CodeProject) working hard but why?

Post by Ferguson »

Using CodeProject 2.2.4 beta, YOLOv5 6.2 1.6.1, BI 5.7.9.12 on W10x64 Pro.

I recently added a new camera, and as a side effect have been paying some much needed attention to my setup. I set the new camera up (a Reolink RLC-823A 16x) to do the motion detection itself, previously it was all BI. Not sure that is relevant. I record continuously.

At issue is when I watch the CodeProject console I see a lot of "Rec'd request for object detection" that fail. More than I expected.

In BI I have the Trigger, AI set for Fire: Only when confirmed, and Cancelled Alerts: Move to cancelled alert lists. My THOUGHT (apparently wrong) is I would not see non-confirmed items normally but could look them up if desired.

But when I look at the Alerts, cancelled, I do not see any alert near the time stamp when AI was doing object detection. When I watch the screen I do not see any indication of motion, though there are blowing leaves and such.

My concern is not so much that the AI is filtering false alarms, that's great, it is that I cannot figure out where they are coming from. Should I be able to see these as cancelled alerts? If not, how can I figure out where they arise, which camera, etc? I ran the AI log up to debug and no real indication in it of a BI related name (like camera) just a guid-like identifier. Nothing in BI's status display. Since I can't find an alert from BI I can't look at the BI status AI pane.

When I get actual confirmed alerts everything works fine, I have it sending an email (for now) and also sending MQTT to Home Assistant, both work great, the clip appears, I can replay it off the timeline or from the clip browser. All good. I would not think anything is wrong if I did not look at all the unconfirmed ones flowing through the console.

For context I have seven 8mp cameras running at 4fps, CPU utilization is around 30% while the console is up on a headless system, GPU is about 50% (Intel HD graphics card built in), I7-6700K, 64gb. If I drop the console so am not monitoring it the utilization drops to about 15% CPU and 19% GPU. So plenty of resources.

What am I missing?

Linwood

Image
Ferguson
Posts: 33
Joined: Sat Feb 22, 2020 1:32 am

Re: AI (CodeProject) working hard but why?

Post by Ferguson »

I decided to turn off cameras to find out which one(s) were causing it, and the answer appears to be "all of them". With all seven cameras disabled the flow of requests stops, but with any camera active it continues, both cameras where BI is doing the motion detection and the camera is doing it.
Ferguson
Posts: 33
Joined: Sat Feb 22, 2020 1:32 am

Re: AI (CodeProject) working hard but why?

Post by Ferguson »

I think I can answer my own question and it is a bit obvious, and I should have read the manual better. I had the (default) per camera AI option for "static Object Analysis" set, which causes a regular request for analysis even if no motion occurs.

I'm not sure that is bringing any benefit to me (though the interaction of that with moving objects is still a bit vague to me), but turning it off removes this phantom load.
User avatar
TimG
Posts: 2098
Joined: Tue Jun 18, 2019 10:45 am
Location: Nottinghamshire, UK.

Re: AI (CodeProject) working hard but why?

Post by TimG »

Static - you got it !
Forum Moderator.
Problem ? Ask and we will try to assist, but please check the Help file.
Post Reply