Dahua IVS deep-learning NVR Wizmind

michele5353

Young grasshopper
Jul 10, 2023
83
11
italy
Hello everyone.
I currently have a dahua system with 24 cam (8mp wizsense) and an nvr5232-el wizsense.
My goal is to be able to carry out a more high-performance AI analysis with IVS, so that I can integrate my cams into my anti-theft system (therefore use them as external detectors).

I can't find much info online but it seems that the wizsense series performs lower level ivs analysis than that of a wizmind nvr (with deep learning based ai). Precisely for this reason I would like to replace my current NVR with an NVR608H-32-XI and have the NVR perform the analysis of some of the 24 cams (those that protect the perimeter).

My questions are as follows:
Do any wizmind experts/users confirm that the analysis of the wizmind NVR is significantly superior to that of wizsense?

I also read that from the datasheet you can use ivs in "heterogeneous" mode, what is this referring to?

In the event that this NVR is not very suitable for obtaining a reliable AI analysis, advice on another type of hw is welcome.
 
Keep in mind that WizSense and Wizmind are simply marketing terms.

WizSense is geared towards the smaller installations (and usually cheaper cameras) and WizMind is geared towards businesses (usually more expensive cameras) - so things like people counting, loitering, etc. are important to them (although some of the 5442 (54IR) series has those as well).

You need to look at each camera and figure out which one you need and how many you want to do AI with. So you may have to double the number of channels for the NVR than you have number of cameras to do what you want.

Having the NVR do the AI instead of the camera cripples the capacity of the NVR and some parameters by half.

For example the 8 channel NVR8CH-8P-2AI is only capable of doing AI on 4 cameras, not all 8. So if you have 6 cameras you want the NVR to do AI on, the 8 channel NVR won't cut it.

And if you use the AI in the NVR, then the bandwidth is cut tremendously.

1735583661278.png




1735583684923.png




What exactly do you mean by "high-performance AI analysis with IVS"?

If you mean triggering for a person or vehicle and nothing else, that is how most of us have the system set up.

Tell us your use case and expected goal and we can tell you if it is possible and with what gear.
 
  • Like
Reactions: bigredfish
Keep in mind that WizSense and Wizmind are simply marketing terms.

WizSense is geared towards the smaller installations (and usually cheaper cameras) and WizMind is geared towards businesses (usually more expensive cameras) - so things like people counting, loitering, etc. are important to them (although some of the 5442 (54IR) series has those as well).

You need to look at each camera and figure out which one you need and how many you want to do AI with. So you may have to double the number of channels for the NVR than you have number of cameras to do what you want.

Having the NVR do the AI instead of the camera cripples the capacity of the NVR and some parameters by half.

For example the 8 channel NVR8CH-8P-2AI is only capable of doing AI on 4 cameras, not all 8. So if you have 6 cameras you want the NVR to do AI on, the 8 channel NVR won't cut it.

And if you use the AI in the NVR, then the bandwidth is cut tremendously.

1735583661278.png




1735583684923.png




What exactly do you mean by "high-performance AI analysis with IVS"?

If you mean triggering for a person or vehicle and nothing else, that is how most of us have the system set up.

Tell us your use case and expected goal and we can tell you if it is possible and with what gear
hi, thanks for the answer
It's written exactly what I have to do. I specified that with my current system (ai by camera with wizsense cameras) the ivs is not 100% reliable enough to be able to integrate it with an anti-theft system (with sirens and alarms).

The question is clear: is wizmind's ivs analysis more powerful? Has anyone already done a similar implementation achieving a reliable level of analysis?
 
In that case, it won't get any better results using the AI in the NVR.

Again Wizmind is just a marketing term. The AI is the same for detection and uses AI IVS whether is is a Wizmind or Wizsense model. Wizmind just adds a few more bells and whistles most don't use.

Which cameras do you have - many of the 8MP models like the 3849 are known to be poor with false and no triggers.

Or maybe you are trying to do too much with one field of view.

Many of use here have our systems set up with sirens and alarms and alerts if someone triggers the camera with little to no false triggers. Most here are doing that with the 5442 series and lower cameras.

But most here are not going to suggest having a camera system replace an alarm system either.

As good as I have my camera system dialed in to only trigger for events I want, I still have an alarm system as well. As do most here.
 
In that case, it won't get any better results using the AI in the NVR.

Again Wizmind is just a marketing term. The AI is the same for detection and uses AI IVS whether is is a Wizmind or Wizsense model. Wizmind just adds a few more bells and whistles most don't use.

Which cameras do you have - many of the 8MP models like the 3849 are known to be poor with false and no triggers.

Or maybe you are trying to do too much with one field of view.

Many of use here have our systems set up with sirens and alarms and alerts if someone triggers the camera with little to no false triggers. Most here are doing that with the 5442 series and lower cameras.

But most here are not going to suggest having a camera system replace an alarm system either.

As good as I have my camera system dialed in to only trigger for events I want, I still have an alarm system as well. As do most here.
I have several DH-IPC-HFW3841E-S-S2 and some IPC-HFW2841T-ZAS.

Unfortunately the IVS they have on board seems not to be 100% efficient.
If what you say is true (so wizmind is the same as wizsense) what type of hw should I use to improve the reliability of the ai? For example dahua IVDs?
 
Both of those models are 8MP on sensor designed for 2MP, which means that little sensor has to work over 4 times as hard than on the 2MP. So it is easier to max out the processor and it misses a trigger.

And without seeing an example of a false trigger, the other likely cause is trying to do too much with one field of view.

At night a 2MP will outperform the 8MP on the same size sensor.

Many here feel 4MP is the go to resolution for these types of cameras.

Many feel the 5442 series is the best overall bang for the buck. It is 4MP on the larger and ideal 1/1.8" sensor.

I have found the AI of the cameras to work even in a freakin blizzard. My only trigger on that camera that night was for the person and not the snow.

1679354257954.png


Or another camera and not one false trigger by the snow.

1733966079326.png



View attachment Snowstorm.mp4

Most of us here rarely got a false trigger using AI with IVS rules.

See this thread on how using just Dahua AI may be sufficient for your needs (and other cameras with AI would perform similar if they are on ideal MP/sensor ratios and you try not to do too much with one field of view):

Who uses Dahua AI capable cameras? Reliable AI for triggering events? Pro's/con's?
 
  • Like
Reactions: bigredfish
Both of those models are 8MP on sensor designed for 2MP, which means that little sensor has to work over 4 times as hard than on the 2MP. So it is easier to max out the processor and it misses a trigger.

And without seeing an example of a false trigger, the other likely cause is trying to do too much with one field of view.

At night a 2MP will outperform the 8MP on the same size sensor.

Many here feel 4MP is the go to resolution for these types of cameras.

Many feel the 5442 series is the best overall bang for the buck. It is 4MP on the larger and ideal 1/1.8" sensor.

I have found the AI of the cameras to work even in a freakin blizzard. My only trigger on that camera that night was for the person and not the snow.

1679354257954.png


Or another camera and not one false trigger by the snow.

1733966079326.png



View attachment 213429

Most of us here rarely got a false trigger using AI with IVS rules.

See this thread on how using just Dahua AI may be sufficient for your needs (and other cameras with AI would perform similar if they are on ideal MP/sensor ratios and you try not to do too much with one field of view):

Who uses Dahua AI capable cameras? Reliable AI for triggering events? Pro's/con's?
I don't quite understand what you mean but I imagine you mean that the cams I have are a bit "underpowered" for what I want to do, right?
 
  • Like
Reactions: bigredfish
Yes, you have budget cams that put high MP onto a sensor designed for much less MP, so the camera has to work harder than what is was designed for and in some instances that can be problematic.

Like trying to take a Ferrari to tow a 40,000 pound trailer. It has the horsepower to do it, but it wouldn't be very good at towing a trailer across the country.
 
Yes, you have budget cams that put high MP onto a sensor designed for much less MP, so the camera has to work harder than what is was designed for and in some instances that can be problematic.

Like trying to take a Ferrari to tow a 40,000 pound trailer. It has the horsepower to do it, but it wouldn't be very good at towing a trailer across the country.
ok, thanks for the advice. to understand better so in your opinion it makes no sense to look for something that can perform a more accurate analysis but would it be better to replace the cams with a more appropriate model? if so, which one?
 
  • Like
Reactions: bigredfish
Yes. Getting a newer NVR and trying to use it to do the AI will result in crippling the capacity of the NVR. And many have shown the AI within the camera is better than the AI done by the NVR.

The 5442 series which is 4MP on the 1/1.8" sensor.
 
Yes. Getting a newer NVR and trying to use it to do the AI will result in crippling the capacity of the NVR. And many have shown the AI within the camera is better than the AI done by the NVR.

The 5442 series which is 4MP on the 1/1.8" sensor.
ok thanks for the advice. I will try to delve deeper and in the meantime if any other user has anything else to say it is a pleasure to compare.
 
Agree 100% with @wittaj

Your problem isnt the NVR and wont be fixed by another NVR. (btw the NVR6 series are way behind on firmware development from the 5 series you already have)

The fix is to get better cameras. IVS/Ai at the "edge" at the camera will most always be superior to doing it on the NVR
 
Agree 100% with @wittaj

Your problem isnt the NVR and wont be fixed by another NVR. (btw the NVR6 series are way behind on firmware development from the 5 series you already have)

The fix is to get better cameras. IVS/Ai at the "edge" at the camera will most always be superior to doing it on the NVR
@bigredfish @wittaj
thanks to both of you for the advice. I have never given too much weight to the mp/sensor ratio. My perimeter sides are about 40 meters long and there are currently 2 cams on the two opposite sides that "look at" each other. based on your advice I thought of the following arrangement: main side 2 ipc (8 mp with 1/1.8 DH-IPC-HFW7842H-Z sensor which is not brand new, it was from the ultra ai series and DH-IPC-HFW5842HP-ZHE). side 2 I have some bf1241 thermals that work very well. side 3 and side 4 I currently have the hfw2841-zas that I would like to replace with one with a more performing optics (can you recommend a 4mp and 8mp model?). ps: regarding the hfw3841 instead, could they be good for performing ivs over short distances? like 10 meters? Thank you!!