Blue Iris

On my doorbell camera, I want to setup a rule that performs a specific action only when it detects a person in one zone and a package in the other zone. AI routinely detects the base of the pillar on my porch as a package so I have removed that area from all zones. However, AI is not ignoring that area of the image, so it performs the package alert any time there is a person on my porch, regardless of whether they have a package.

Zone A is the person area
Zone B is the package area
No zones cover the area I want to ignore

AI alert config for package: person:A+package:75B

Static object detection prevents repeated triggers and unnecessary AI analysis, but it doesn't seem to be ignoring the object when there is confirmed objects in other zones.

What am I doing wrong?

Surely the zones only affect triggering and you haven't masked out any area so the AI will look at the entire inage. As a start I think you will need to clone the camera in BI5 and separate your person and package that way. The base of the pillar needs to be masked out - see "blackout" in Help file.

Is there a setting for 'goblins'?

I'm afraid you will have to train your own model for that. You will need a considerable number of pictures to make it accurate

TimG wrote: ↑Sun Sep 22, 2024 10:06 am Surely the zones only affect triggering and you haven't masked out any area so the AI will look at the entire inage. As a start I think you will need to clone the camera in BI5 and separate your person and package that way. The base of the pillar needs to be masked out - see "blackout" in Help file.

Is there no way to tell AI to ignore a portion of the image without creating a blackout mask? I don't want the live view to be blacked out in those areas; I just want AI to ignore it.

Can't you just establish zones and then specify object requirement triggers within the zones?

No zone, no trigger. Seems simple enough.

That's how I do it..., the old fashioned way without all this AI smart stuff providing so many wonderful new features, like identifying a mouse as a mailman and a fence post as a school bus.

How did we ever live without it???

Pogo wrote: ↑Mon Sep 23, 2024 6:13 pm Can't you just establish zones and then specify object requirement triggers within the zones?

No zone, no trigger. Seems simple enough.

That's how I do it..., the old fashioned way without all this AI smart stuff providing so many wonderful new features, like identifying a mouse as a mailman and a fence post as a school bus.

How did we ever live without it???

I have 2 zones created: 1 for where the person is expected to be, and 1 for where I want to detect packages. The base of the pillar is not included in a zone, but AI is still analyzing that area of the image and incorrectly detecting that it is a package. The zones are just for motion triggering, and are apparently not utilized by AI. I'm not interested in simply knowing when there is motion in either/both of the zones. I want to know when there is a person and a package detected at the same time. AI is capable of doing that, but my current solution is worthless because it isn't ignoring the base of the lamp post.

If I enable blackout for masked areas, it should ignore the base of the pillar, but then I have a large black box on my live feeds all the time, which is not ideal. I'll go that route if i need to, but I"m hoping there's a way to tell AI to ignore that specific area of the screen.

dodgeboy wrote: ↑Mon Sep 23, 2024 6:20 pm ...my current solution is worthless because it isn't ignoring the base of the lamp post.

Sort of my point. So is it too dumb, or too smart for its own good? There has to be a simple and effective solution to achieve what you're attempting. Just hafta find it..., and you will.

But isn't it ironic that there was literally no such thing as a 'false positive' until AI came along to eliminate them? Funny how that worked.

An alert is literally a positive response to a trigger any way you want to cut it or define it. It is and never has been a false or negative response to a trigger or there would not have been a trigger to effect a response at all.

Unwanted alerts are just that, unwanted. They're still positive results to a pre-defined trigger that have been inaccurately termed "false positives" -- compared to what? Until AI came along, nothing. There was either a useful alert, or a nuisance alert..., both positive results of preset parameters to create the action.

So as a desperate and totally inaccurate term looking for an identity, along comes AI to the rescue providing absolute (or not) definition for the results of such resulting triggers. Like magic, alerts were something now, (sorta like Navin Johnson in 'The Jerk'). LOL


So until AI, 'false positive' was nothing more than an oxymoron phrase coined to describe the frustration of what were more accurately just undefined positive nuisance alerts.

Now they have names..., in many cases wrong ones with completely inaccurate descriptions or definitions. They are now literally False Positives compared to what the predetermined configurations were designed to produce from a totally positive trigger and subsequent positively expected accurate alert.

"It's a bird!" "It's a plane!" "It's a what?" "A groundhog?" Oops.

AI may eliminate the majority of unwanted nuisance alerts, but it has actually given legitimate life to the term 'False Positive' which technically didn't exist before it.

No reflection on you, Dodgeboy. Just a short rant I've been wanting to vent when an appropriate opportunity came along. This seemed like a good spot to sneak it in without creating too much stress for the AI masses. It just amazes me how much time, energy and money people spend on AI solutions to what in many cases are simple tasks easily handled by the basic BI tools and features.

dodgeboy wrote: ↑Mon Sep 23, 2024 5:56 pm Is there no way to tell AI to ignore a portion of the image without creating a blackout mask?

Create another zone for the area to be ignored and tell AI to do so?

How do you tell AI to ignore a zone? This sounds like exactly what I'm looking for, but I don't know how to do it.