Trigger on bottom of objects entering zone + make BI geometry aware

[vertigo]

Consider this situation, my driveway:

.

Im trying to detect a car or human entering my drive way. This is surprisingly difficult. Because I have to ignore anything driving on the road, so I created this zone:



I want to trigger on any object entering that zone from the road. But a car that actually enters my driveway, especially a large van, will also cover the road. The center of the object may even be "on" the road if its a tall delivery van or truck, or a human close to the camera. So this does not trigger:



If I disable the zone crossing requirement, I need to set a really small object size to fit in the narrow (vertical) zone that is not overlapping with the road, and as a result cats and birds and whatever will cause false alarms.

What would help tremendously is having an option to take only the bottom or bottom 20% of the object in to account to determine if something crosses a zone. I could then use a large object size and trigger on (the bottom, ie wheels) crossing my driveway zone. This could also help in general to determine if a human walks on the road or on the driveway, regardless if he is closer or further away.

Less trivial to implement, but possibly useful would be to make BI aware of world geometry. With that I mean, allow us to configure how the ground plane is oriented in the camera view. That way, you could say: detect objects larger than say 1 meter instead of x pixels, and it could work regardless if the object is in the distance or closer to the camera. At least if you assume the bottom of the object touches the ground (not gonna help with spiders and insects).

An illustration:


I would like to detect humans near my shed, but those would be smaller in pixels than a cat in the foreground. Its not trivial to make this work in BI right now, its doable I think with cloning the camera and creating 2 or 3 sets of zones and triggers, but its really clumsy. If BI understood the bottom of the screen is much closer than the top, it could scale the object size trigger accordingly.

[vertigo]

Im not interested in paying a subscription to solve a problem this trivial. Im not asking to recognize faces (which shouldnt even be very hard using opencv), just detect an object that is 100s of pixels large!

For now I solve it using another DVR app that has no problems differentiating between tall objects in the foreground and objects in background (and that even does recognize vehicles and people and animals), so I use that app to trigger an MQTT via home assistant. It works, but its pretty insane. I thought BI was an upgrade to that aoo, in this respect at least, it certainly isnt.

[vertigo]

Sorry, my bad, I conflated it with Sentry AI. AITools might work.But Ive meanwhile began experimenting with Frigate on a raspberry, and its looking promising, so Im considering moving everything over to that Pi using shinobi or something else.

[MikeBwca]

Your 'Zone A' should detect ANYTHING in the zone if your sensitivity settings are adjusted. Also, the top of your zone is too high. You don't need it to follow the top of the bushes - that is, unless your expecting a semi trailer or Boing 757. When the wind blows and those bushes start to sway, you'll get a lot of triggers.

Can you post captures of your 'trigger' 'motion sensor' settings?

Your using terminology 'Entering the zone'! A zone will detect within the zone, not necessarily entering/exiting the zone. You'll need to use 2 or more zones, and zone-crossing if you want to do this.

[vertigo]

Your 'Zone A' should detect ANYTHING in the zone if your sensitivity settings are adjusted.
(.._
Your using terminology 'Entering the zone'! A zone will detect within the zone, not necessarily entering/exiting the zone. You'll need to use 2 or more zones, and zone-crossing if you want to do this.

Well that is the problem. right now, because of the issue explained above, I am triggering only in zone A (ignoring the rest and not triggering on entering zone A), but that means I only have a very narrow window, in both time and height. A human walking up to my door, especially on the left side (from our perspective, so up) will go through a detection zone that is only from his feet to his knees for most of the time, and depending on clothing a person is often detected as 2 objects that are even smaller. If I set the detection size that small, cats and birds and rodents trigger it too. The other issue is time. If I set the object size larger, humans will walk out of the camera within a second or so and cars that stop on my driveway, I often have less than a second between the object being detected and stopping, so my travel distance has got to be pretty much zilch, my detection time has to got be nearly zilch and the size has got be small to catch humans. That combination just doesnt work.

It would be much better if I could track the objects on the road and trigger upon entering the driveway, which I cant right now for reasons explained above.

Also, the top of your zone is too high. You don't need it to follow the top of the bushes

Thats for people walking past the camera. Bushes arent a problem.

Can you post captures of your 'trigger' 'motion sensor' settings?

Ive tried them all. Ive run recordings through all possible combinations. They cant solve the issues Ive explained above. triggering on the bottom of the object would solve 99% of my problems.

[MikeBwca]

Yes, I understand. The same detection settings do not work for near and far areas. What I've done is to clone the camera, and name it 'Front Door', and set the motion detection for each differently to detect in those specific areas. Use the 'Clone Master' as your driveway detection camera. You'll get better results.

[vertigo]

Yes, I understand. The same detection settings do not work for near and far areas. What I've done is to clone the camera, and name it 'Front Door', and set the motion detection for each differently to detect in those specific areas. Use the 'Clone Master' as your driveway detection camera. You'll get better results.

Its not just a matter of near and far. That would be nice to have, but my main problem is that I can not use zone crossing because BI only considers the center of the object, which may never enter the zone Im interested in, like the grey van in my first past. Seriously this is a no-brainer: the bottom of an object is what matters (for anything that is not flying) and it should be possible to trigger on that. My previous NVR had this option and it worked perfectly.

To use a small hyperbole, I could probably defeat most BI security camera's that use object zone crossing by carrying a tall plank!

[Thixotropic]

It may be that adding a single, dedicated camera with a carefully selected FOV could resolve this. Set it to cover only the particular area of interest and then adjust the motion masks to trigger as needed.

Similar to your situation, I need to trigger on movement within a very specific area within a zone. I have a cam I'll be mounting outdoors, quite a ways from our house. It'll view the entire side of the house but will be set to trigger only on a specific area.

There's no practical way for me to cover the area of interest with the other cameras and zones- it just became unwieldy to arrange the zones and triggers and everything. The dedicated camera will solve this problem 100% and I'll be able to rely on the alerts as being legitimate.

Not the simplest or cheapest solution but it may be the most reliable way to detect the motion/events that you need.

[vertigo]

I cant mount any outside camera's there for.. reasons. Instead, pending I get some sort of tensor flow object recognition working, I have a mounted a PIR sensor to my letter box, which sends an RF signal that I receive with a sonoff bridge, which feeds it in to HA which sends an MQTT to BI and starts recording. Works fine - except in summer when its too hot and the PIR sensor sends non stop signal. Not sure yet how that will work when its freezing.

But that doesnt mean triggering on bottom of an object isnt something that should be added to BI, it can not be hard to implement and its going to help a lot of people in a variety of situations. IN fact I cant think of a situation where the center of an object would be more meaningful than the bottom (ie, ground).

[MikeBwca]

I'd suggest contacting support. Maybe Ken can help.