Review-AI NVR 4216-I

[teemu81]

Based on my experience with the recorder you'll always need a device with FD capabilities to get the face recognition to work. In my tests I've used an NVR4208-8P-I, which shares firmware with the NVR4216-16P-I and the device worked as it should with face detecion cameras, the exact model we used for testing was the IPC-HDW4431EM-ASE. Since it's main feature is the face recognition we focused all our tests on that feature, and everything worked as intended, performing the face recognition with a high rate of success even when the angle wasn't the best one.
As far as I know you can't make it work with regular cameras, at least with the firmware version V4.000.0000000.1.R.20190809 which is the last we tested

 

[c hris527]

Yes, I’m certainly struggling. The fact that I’ve found two data sheets, which appear to show the same model numbers but differing face recognition and ANPR features leave me very confused.

Its NO surprise that USA and International Dahua Differ. Sometimes they differ slightly or sometimes we find out after the fact they differ Greatly. The majority of the time the International ones are very suffice but with all these new AI features being bundled into a consumer end or middle of the road NVR, IMOHO I think things are backsliding for them a bit especially with the firmware issues as of late, fix 2 things, break 1..lol..Overall its still good bet for the price point.

 

[looney2ns]

And at times, the spec sheets are just flat out wrong.

If I get the chance I will give it a go with a standard camera and face detection.

 

[craigspc]

Thanks for all your responses!

 

[mark_whocares]

@looney2ns did this NVR get firmware updates to fix any of your issues ?

 

[redfive]

I too am interested on this, a friend of mine is planning a new install with AI cameras, but I'm still not secure if recommend him the 5XXX -4KS2, or with this AI model ....

 

[Arjun]

For now, I would still recommend the 5XXX -4KS2 models; they support the new NVR 4.0 AI Firmware

 

[redfive]

Yes, I well know the 5XXX-4KS2 model (I've some of 'em around), and the Pro AI is too expensive .....

 

[looney2ns]

@looney2ns did this NVR get firmware updates to fix any of your issues ?

No, not yet.

 

[forinsec]

Thanks guys for your extensive testing, I will stick to the 5XXX -4KS2 models after reading through all of this

 

[m00st]

And at times, the spec sheets are just flat out wrong.

If I get the chance I will give it a go with a standard camera and face detection.

Hi Looney did you ever get to play with the face recognition on a non-FD camera?

Update May 30th: face recognition works with any camera for up to 2 cameras. Is does not work in conjuction with IVS or heatmap.

 

[m00st]

I can't hear it 5ft away. So far, it's running fine. It's been running roughly 4 weeks, continuously at this point.

I could definitely hear the fan of my 4216-AI while the NVR was at about 3ft from me for the past few weeks. I got myself a nice Noctua 40mm fan which should be a lot quieter. By itself the fan is inaudible, but as soon as you close the case , you hear the airflow. Still considering it a worthwhile upgrade, less noise. One word of caution: the polarity on the motherboard and fan cable are wrong. Red is negative and black is positive. Good thing that Noctua fans have a diode.

 

[jmcu]

I just picked up a A4x10 that I am going to try out with the 5208-4ks2. Why did you choose the thicker - x20?

 

[m00st]

Hi Looney did you ever get to play with the face recognition on a non-FD camera?

I just picked up a A4x10 that I am going to try out with the 5208-4ks2. Why did you choose the thicker - x20?

it was the one they had in stock

 

[ArnonZ]

[QUOTE="but the Dahua IPC-D5442E-ZE (pro AI model) looked quite interesting
[/QUOTE]
I too was looking for the combination of Dahua IPC-D5442E-ZE/(IPC-HDW5442E-ZE) together with either 4216-16P-I or the 5216-16P-I
while license plates are not too important for me, the face recognition is (even though the height of the cams in my setup is not ideal).

• Does anyone have already some experience with that combination or with anything close to it (any of the 5442 series)?
• Any news about the facial recognition capabilities?
• Maybe @looney2ns that was playing with the setup knows if I can (in theory at least) use some sort of combination of tripwire+human+stranger to trigger the alarm?

I do expect that in such scenario, if it works, I may still have some false alarms if the camera identified a tripwire of human but did not identify the person and therefore think it's a stranger.
It's still better than the case of alarming each time a human crosses a tripwire (in theory that is).

The last question would be about multiple cams observing the same person: in ideal world, I'd like to choose that if I was identified by one of the cams only, the alarm would not set off.
But I guess that in practice, the second cam that did not identify me would not be able to tell that the stranger is me. it could still be nice if there was anything close to ITTT so that I could choose not to trigger the alarm, by any cam, X minutes after a family member was identifies in my property. I am aware that in such setup a burglar can break in while I an sitting with my wife at the back yard but that is something I can live with.

I know that I'm expecting the AI features that the AI NVR (supposed to) have together with the "brains" of the BI but still, it's worth asking. Also, if there is any other setup that is different that what I've had in mind but solves my problem, I'd be happy to hear about it.

 

[ArnonZ]

from what I've red so far about the 4216-I, it looks like too big of a risk to buy it.
I was thinking about a set of 8XHDW5442T-ZE together with NVR5216-16P-I
Andy suggests that the 5216 would be better if only because of the fact that I intend to connect 8IPC with FD to it.

Before pulling the trigger, I wanted to ask you guys if anyone have some experience with the combination and/or if you guys have any remarks/insights.
I would like to try and use the 8 cams to trigger the alarm given human+cross tripwire + stranger.

 

[gecko]

Just curious, after reading this thread I can assume that a Dahua NVR is necessary for face detection? Is there any standalone camera that can do this so the NVR isn't necessary? I have a milestone setup running currently and don't really want to have to redo the entire deal.

 

[teemu81]

Last time I've checked the prices for the face recognition cameras (which exist) were over the top, it was way cheaper to set up one of these NVR for the face recognition.

 

[gecko]

Something to think about then. This whole covid thing is making people crazy. Crime is way up

 

[ArnonZ]

Just curious, after reading this thread I can assume that a Dahua NVR is necessary for face detection? Is there any standalone camera that can do this so the NVR isn't necessary? I have a milestone setup running currently and don't really want to have to redo the entire deal.

There are face recognition cams that can do it without the need of a specific NVR and those can either use them, can use several ordinary cams OR can get FR for more cams using cams that support FD (face detection): the NVR can use the face captures form the FD cams and use it for face recognition.
In theory, if Dahua took it seriously, the NVR way is much better: you want the DB of the recognized faces to be shared by all cams and you want each face you've added to the DB to be known by all cams.
BUT
Befor you're thinking about face recognition, you must remember few facts: FR is not and cannot be as reliable as other methods. To get a good identification, you should also keep a short distance, the right light and the right angle and even then, not every capture would be good enough.
When a face IS recognized, you can trust it but not always would it be recognized. Meaning I've never got a false positive but false negative is very common.
Saying that, it had to be implemented differently:

• Just like I can limit the area of detection for IVS and for SMD, for FR it's even more necessary but it's not implemented unfortunately.
• Since FR cannot be as reliable as other ways, it could be good only as an extra filter for IVS and/or SMD. That way, you could also solve the problem of area limitation (two in one). But they didn't do it.
• Since not all face captures are good enough and since the problem is with false negative only, the right implementation had to be to follow the same intruder for some time, to keep taking captures and if one of the capture matches within X seconds (or till the person moved out of frame), you can make your decision about detection.
• The DB had to be dealt as in AI: all of my positive detection had to be saved under the same DB in order to "know" me better and improve my identification. it's not happening that way. All of my pictures are saved at the DB as different person with the same name. There is no learning and that's NOT AI.

Those 4 issues are all software issues that could be fixed but they weren't. They could turn a completely useless feature to be the ultimate machine.
The potential is great, the HW is good enough and it's all doable.
For now, the cold truth is that the FR NVR is no more than a stupid gimmick that is good for NOTHING.