Dahua Mini-PTZ (SD1A404XB) Smart Plan Setup **SOLVED**

[CanCuba]

Been playing with this camera for a few days and I really like it. Responsive enough despite not being a speed dome. Small form-factor which I love and very subtle. I like that it's almost impossible to make out where the lens is pointing.

Let me begin by setting out why I bought this camera and what I want to achieve:

• primarily, face detection at my front entrance and of the considerable foot traffic on the sidewalk and street
• camera stays in "Overwatch" preset all day (camera isn't running at night as it's on a temporary rig)
• camera is triggered to different presets by tripwires set up on other cameras ("overview cameras")
• no touring, panning, patterns, etc.

My questions are regarding the Smart Plan:

1) I can assume that the Global Setting FD is to be left on as the primary purpose of this camera is for FD, yes?
2) If I set a preset (say 1: Overwatch), to Deep IVS or people counting, does this disable FD, while in the said preset, in favour of IVS?
3) Referring to question 2, could I set Overwatch to Deep IVS and then use the tripwires to trigger the camera, via the NVR, to go to another preset?
4) Or, should I just leave all the presets in FD?

Dahua documentation is thin, to say the least, and I haven't found any YT videos covering this.

Any help is greatly appreciated!

 

[wittaj]

Are you finding that face detection actually works?

Yes, you can only run one feature at a time.

Yes you can set up a spotter cam in the NVR to go to any other preset.

 

[Mark_M]

1) I can assume that the Global Setting FD is to be left on as the primary purpose of this camera is for FD, yes?

When global is set to Face detection, it will be performing face detection when the camera is not at a pre-set.
Such as manual control as you look around.

2) If I set a preset (say 1: Overwatch), to Deep IVS or people counting, does this disable FD, while in the said preset, in favour of IVS?

Yes, only one thing can be enabled at a time.
A pre-set takes priority over global plan.

3) Referring to question 2, could I set Overwatch to Deep IVS and then use the tripwires to trigger the camera, via the NVR, to go to another preset?

Yes, your NVR should pick up the pre-sets of the camera and be able to configure it.
It can use other cameras to trigger this camera to a pre-set.

4) Or, should I just leave all the presets in FD?

Face detection has it's useses.
It is NOT face recognition.
All the camera is doing is finding what looks to be a face then performing an action (siren, snapshot, etc).

For an overview camera looking at a driveway, or similar, it can be useful to search for any person detected during the day.
Such as searching when a person was around but didn't trigger an IVS rule.

 

[Timokreon]

I use face detection on my side/porch camera, 4k-x. Gives me a snapshot of people coming up to my porch, or trying to get into my side yard. It occasionally picks up one of my bushes as a face when it's windy out, but not often.

 

[wittaj]

I use face detection on my side/porch camera, 4k-x. Gives me a snapshot of people coming up to my porch, or trying to get into my side yard. It occasionally picks up one of my bushes as a face when it's windy out, but not often.

Curious why you don't run human IVS on the 4K-X? With face detection, if someone is hiding their face or has a hoodie on or looking down, the face detection could miss them, while the human detection would still get them.

 

[Timokreon]

I have two other cameras covering the same general area. That and I just wanted to see if the face detection worked for that spot.
I also plan on putting the mini-ptz at the front of the porch for tracking, hence eventually ending up with 4 cameras covering the basic area.

 

[CanCuba]

Are you finding that face detection actually works?

Yes, you can only run one feature at a time.

Yes you can set up a spotter cam in the NVR to go to any other preset.

Yes, I'm getting about 120 face detections a day in an area with moderate foot traffic. This is with the camera mounted on a "testing rig" (attached to a board which is strapped horizontally to a ladder on my 2nd floor terrace overlooking the street) which is restricted in its coverage area. When I have it properly mounted, the camera will have about twice as much view.

I bought this camera for FD as the rest of my property is covered. It's been useful as the neighbours and I are using it to figure out who the local thieves are. In the first two hours on its testing rig, the camera (as well as my other cameras) caught someone sneaking around. The same individual ended up stealing an unattended mobile phone! A tripwire was activated on a spotter cam, PTZ moved to the preset and got a perfect face capture. Cops loved the face capture as well as the 13 videos that the same individual appeared on!

 

[CanCuba]

When global is set to Face detection, it will be performing face detection when the camera is not at a pre-set.
Such as manual control as you look around.



Yes, only one thing can be enabled at a time.
A pre-set takes priority over global plan.



Yes, your NVR should pick up the pre-sets of the camera and be able to configure it.
It can use other cameras to trigger this camera to a pre-set.

View attachment 145386


Face detection has it's useses.
It is NOT face recognition.
All the camera is doing is finding what looks to be a face then performing an action (siren, snapshot, etc).

For an overview camera looking at a driveway, or similar, it can be useful to search for any person detected during the day.
Such as searching when a person was around but didn't trigger an IVS rule.

View attachment 145387

Thank you. This camera's sole function will be FD. I thought maybe I could set up some more tripwires and keep FD but doesn't appear to be the case.

 

[CanCuba]

I use face detection on my side/porch camera, 4k-x. Gives me a snapshot of people coming up to my porch, or trying to get into my side yard. It occasionally picks up one of my bushes as a face when it's windy out, but not often.

I dropped the quality threshold as low as it will go and set to instant snap. I get some interesting things detected as faces (bushes, shadows, wheel rims, etc) but I've also gotten some faces as a surprising distance for a camera with 12mm max focal length! The camera is about 8m up on it's testing rig and some of the faces detected were over 15m away. It's been a very useful tool so far.

 

[wittaj]

@CanCuba we would love to see some snaps of these captures at distance!

I do not have that particular camera, but FD is just detecting a face and triggering and not facial recognition that you could assign names to faces correct?

 

[CanCuba]

@CanCuba we would love to see some snaps of these captures at distance!

I do not have that particular camera, but FD is just detecting a face and triggering and not facial recognition that you could assign names to faces correct?

Correct, just detection and not recognition.

I'll upload some snaps later in the Pic forum. They're not great but during daytime it's easy to identify the person.

I live in Cuba and everyone knows their neighbours, generally. The national identity cards are digitized so the cops are going to run the face captures I have though the system. They're expecting to get a hit.

Backend face recognition is easy enough. iPhones do it. So once I get a big enough database, I can just let software on my laptop identify the same faces and then put names to the faces. I use Linux and DigiKam does face recognition quite well.

 

[Mark_M]

FD is just detecting a face and triggering and not facial recognition that you could assign names to faces correct?

Yes, and if your NVR supports Face Recognition, then Face Detection also needs to be turned on (usually the camera or NVR allows for both at the same time).

I have 'Video structurization' turned on from my NVR. By default it also does Face detection.
When I search for any people seen, I get a search result like this:
If someone is looking towards the camera it also triggers 'Face detection' and that's the snapshot with some of them.

 

[CanCuba]

Yes, and if your NVR supports Face Recognition, then Face Detection also needs to be turned on (usually the camera or NVR allows for both at the same time).

I have 'Video structurization' turned on from my NVR. By default it also does Face detection.
When I search for any people seen, I get a search result like this:
If someone is looking towards the camera it also triggers 'Face detection' and that's the snapshot with some of them.

View attachment 145517

No face recognition on my NVR. Looking at Frigate NVR which does FR. Will be my winter project.

 

[danbutter]

No face recognition on my NVR. Looking at Frigate NVR which does FR. Will be my winter project.

Just a quick FYI...Frigate does not do face recognition. Frigate does object detection. You can use doubletake with frigate which then uses other things like deepstack or compreface to do face recognition.
You can make it all work together so that if frigate detects a person doubletake can be made aware of this and then leverage whichever face detection you would like to use. All takes a good amount of processing power to accomplish. A google coral (super hard to find right now) is very strongly recommended for this.
If you are already running servers 24/7/365 that have the overhead to do all of this then great. Otherwise getting an NVR or cameras that have this built in is likely a better idea. Especially for your electricity bill.

 

[CanCuba]

Just a quick FYI...Frigate does not do face recognition. Frigate does object detection. You can use doubletake with frigate which then uses other things like deepstack or compreface to do face recognition.
You can make it all work together so that if frigate detects a person doubletake can be made aware of this and then leverage whichever face detection you would like to use. All takes a good amount of processing power to accomplish. A google coral (super hard to find right now) is very strongly recommended for this.
If you are already running servers 24/7/365 that have the overhead to do all of this then great. Otherwise getting an NVR or cameras that have this built in is likely a better idea. Especially for your electricity bill.

I've been watching some videos about Home Assistant/Frigate/Doubletake integration. It's what I'm looking for and just a matter of seeing what kind of hardware I'll need to accomplish what I want to do. 1x 4K camera, 1x 2MP camera, 12x 4MP cameras. All at 15fps 24/7 recording with event notification.

Electrical consumption isn't so much a matter of utility cost for me. I get blackouts (some scheduled, some not) so consumption is more of a concern in terms of my battery backup (12v 150ah gel battery array) being able to handle several hours. If we get a hurricane, it can be up to 5 days without power.

There's another post I can't track down right now with someone running HA/Frigate/etc on a mini pc. Very viable option for me if I can get rid of the NVR. The Dahua NVR has been fine but I'm looking at taking things a set further. BI isn't an option for me as I only run Linux.

 

[danbutter]

I've been watching some videos about Home Assistant/Frigate/Doubletake integration. It's what I'm looking for and just a matter of seeing what kind of hardware I'll need to accomplish what I want to do. 1x 4K camera, 1x 2MP camera, 12x 4MP cameras. All at 15fps 24/7 recording with event notification.

Electrical consumption isn't so much a matter of utility cost for me. I get blackouts (some scheduled, some not) so consumption is more of a concern in terms of my battery backup (12v 150ah gel battery array) being able to handle several hours. If we get a hurricane, it can be up to 5 days without power.

There's another post I can't track down right now with someone running HA/Frigate/etc on a mini pc. Very viable option for me if I can get rid of the NVR. The Dahua NVR has been fine but I'm looking at taking things a set further. BI isn't an option for me as I only run Linux.

Wow. I think that is a tall ask for a mini PC. Most desktop PCs as well. You want to take 58MP 15 times a second, run object detection and then facial recognition and then have home assistant send out notifications after running an automation.
It would take a pretty beefy computer to do all that and you'd need a google coral (or 9ish...I thought they were good for about 100fps, but I could be wrong) and some sort of pretty good GPU to help decode the video streams.
Good luck and let us know how it goes.

 

[CanCuba]

Wow. I think that is a tall ask for a mini PC. Most desktop PCs as well. You want to take 58MP 15 times a second, run object detection and then facial recognition and then have home assistant send out notifications after running an automation.
It would take a pretty beefy computer to do all that and you'd need a google coral (or 9ish...I thought they were good for about 100fps, but I could be wrong) and some sort of pretty good GPU to help decode the video streams.
Good luck and let us know how it goes.

I wouldn't be running facial recognition on all the cameras. 4 at most and not in real-time.

Frigate is designed to use a substram at 720p @ 5fps. So 20 cameras would get to 100 fps for object detection.

For sure, at least one Coral, maybe two.

This is within the capabilities of a mini PC with at least one Coral.