Review - Dahua IPC-HFW7442H-Z 4MP Ultra AI Varifocal Bullet Camera

Wildcat_1

Known around here
Joined
Dec 9, 2018
Messages
2,047
Reaction score
5,867
Location
US
All, quick update. I just spoke with Andy @EMPIRETECANDY and good news, he has more of the standard 7442H-Z’s (featured in this review) coming in on November 19th. He also has some Z4s if anyone is looking.
 

Davahad

Getting the hang of it
Joined
Nov 7, 2016
Messages
126
Reaction score
84
@Wildcat_1 Any ideas on how the "Deep Learning AI" works? For example, I have my IPC-HFW7442H-Z4 focused on a residential street only counting cars using the Video Structuralization (1 Rule = Traffic Flow Stat+ OSD Enabled + Snap Mode Optimized). Not recording plates or taking pictures of cars, counting people, etc.).


Theoretically will the camera get more accurate over time at counting cars? Currently I would guess that it is about 98% accurate counting cars and it misses every once in a great while but not sure why.

Thanks,
David
 

Wildcat_1

Known around here
Joined
Dec 9, 2018
Messages
2,047
Reaction score
5,867
Location
US
@Davahad looks like you’re running SSA there or if not have WDR on ?

With regards to Deep Learning AI that really is speaking to what the camera is already doing such as deep metadata (vehicle brand, plate, color, age of person, expression etc) but as there is no way to calibrate or weight this as a user there is no way to ‘train’ the cam or an AI NVR ‘yet’. One exception is Face Recognition based on the ZFR version of this cam as you can record failures of that in an AI NVR once you’ve populated its database with faces.

HTH

Here is a link to a 2017 Dahua video ad on Deep Learning AI.




@Wildcat_1 Any ideas on how the "Deep Learning AI" works? For example, I have my IPC-HFW7442H-Z4 focused on a residential street only counting cars using the Video Structuralization (1 Rule = Traffic Flow Stat+ OSD Enabled + Snap Mode Optimized). Not recording plates or taking pictures of cars, counting people, etc.).


Theoretically will the camera get more accurate over time at counting cars? Currently I would guess that it is about 98% accurate counting cars and it misses every once in a great while but not sure why.

Thanks,
David
 

Davahad

Getting the hang of it
Joined
Nov 7, 2016
Messages
126
Reaction score
84
Thanks for the catch on WDR. Was initially testing a bunch of settings Day/Night and forgot to turn that off.

The video mentions that accuracy of pattern analysis & classification are increased over time. This seems to imply it does it on its own but not sure how that is possible without user feedback showing cars it did not count that it should have. Perhaps its leading to what you described about providing feedback on the facial recognition?
 

Davahad

Getting the hang of it
Joined
Nov 7, 2016
Messages
126
Reaction score
84
Hopefully in a future update they will add the ability to train the camera based on the scene, etc.
 

Wildcat_1

Known around here
Joined
Dec 9, 2018
Messages
2,047
Reaction score
5,867
Location
US
that is my thought, yes. Again without user input or correction the AI can only do limited AI learning using its own SOC processing and that would be limited to comparison against any internal algorithms on the phone. Therefore ‘training’ would be key to overall improvement of the AI. The good news is even if the camera is limited in that way (say this is classed as V2.0 AI for Dahua and V3.0 releases whenever that would feature training / calibration) that it’s conceivable that when paired with an AI NVR, the NVR could be used as the training agent that when images are passed back from the cam, the NVR learns and improves. That way we have options in that regard too.

Thanks for the catch on WDR. Was initially testing a bunch of settings Day/Night and forgot to turn that off.

The video mentions that accuracy of pattern analysis & classification are increased over time. This seems to imply it does it on its own but not sure how that is possible without user feedback showing cars it did not count that it should have. Perhaps its leading to what you described about providing feedback on the facial recognition?
 

Davahad

Getting the hang of it
Joined
Nov 7, 2016
Messages
126
Reaction score
84
I think I found a bug. Video Structuralization - Reports - Daily
You cannot do the max range of 24 hours. You can only do 23 hours.
Start Time 12:00:00 AM End Time 11:00:00 PM. I cannot change the minutes or seconds digits to make 11:59:59 PM either.
 

Wildcat_1

Known around here
Joined
Dec 9, 2018
Messages
2,047
Reaction score
5,867
Location
US
yes that is 1 that fed back on :)

I will be writing my summary post over the next few days on overall impressions, findings and recommendations. Will cover FW bugs found then as well. Nice thing is that FW bugs are ‘relatively’ easy to have fixed :)

I think I found a bug. Video Structuralization - Reports - Daily
You cannot do the max range of 24 hours. You can only do 23 hours.
Start Time 12:00:00 AM End Time 11:00:00 PM. I cannot change the minutes or seconds digits to make 11:59:59 PM either.
 

spenser911

Getting comfortable
Joined
Jan 25, 2019
Messages
40
Reaction score
21
Location
USA
Absolutely amazing review - Do you know what algorithm version these ultra series are using?

Why would anyone buy a dedicated Dahua ANPR anymore? When this does that AND much more..
 

Wildcat_1

Known around here
Joined
Dec 9, 2018
Messages
2,047
Reaction score
5,867
Location
US
The Ultra AI really are amazing cams and to my knowledge represent v2.0 of Dahua’s AI platform currently.

Absolutely amazing review - Do you know what algorithm version these ultra series are using?

Why would anyone buy a dedicated Dahua ANPR anymore? When this does that AND much more..
 

Davahad

Getting the hang of it
Joined
Nov 7, 2016
Messages
126
Reaction score
84
yes that is 1 that fed back on :)

I will be writing my summary post over the next few days on overall impressions, findings and recommendations. Will cover FW bugs found then as well. Nice thing is that FW bugs are ‘relatively’ easy to have fixed :)
Would be better if they used 00 to 24 like they do on the NVR
 

spenser911

Getting comfortable
Joined
Jan 25, 2019
Messages
40
Reaction score
21
Location
USA
Wildcat
The Ultra AI really are amazing cams and to my knowledge represent v2.0 of Dahua’s AI platform currently.

Whenever you have a moment - do you mind checking for me? Here is what our dedicated FR camera is on
1574293431734.png
 

spenser911

Getting comfortable
Joined
Jan 25, 2019
Messages
40
Reaction score
21
Location
USA
In-terms of catching plates, how do you feel about the performance with Video Structure enabled? On a 5832 Series Pro AI NVR (doing the processing) the results are less than impressive. From the looks of your pictures it blows this NVR out of the water.
1574360082291.png

1574359923664.png
 

EMPIRETECANDY

IPCT Vendor
Joined
Nov 8, 2016
Messages
8,260
Reaction score
23,722
Location
HONGKONG
My final summary, thoughts recommendations added into post #5
Thanks to @Wildcat_1 fantastic review, during the testing, we talked a lot about this model, and lastly he shows all good and bugs that this camera have. I will keep following on the update for these bugs.
This series right now have these models
IPC-HFW7442H-Z 4mp
IPC-HFW7442H-Z4 4mp
IPC-HFW7842H-Z 8mp
12mp still not release, I think can book next month I think.

We also will have a 8MP Bullet IPC-HFW58XX-ZE AI new cam coming out in Dec, will grab one for testing.
For the AI cams, better use NVR52xx-4ks2 NVR41XX-4KS2 with 4.0 AI version firmware to use,so right now is the important that you have a original Multi-languages NVR. Hacked one never can be updated.

If good budget is good, can use NVR4208-8P-I, NVR5216-16P-I, or some other Ultra AI NVR.
For budget limited users, those famous cams on this forum still be a good choice and can beat down 85% of cams in the world. If use these ultra smart system then can beat down 99% in the world.:headbang:
Not everyone can buy those expensive ones, but we can share the LATEST new tech here for fun, that is my purpose.
I will keep on sending more new cams for reviewing on the forum. Thanks to @fenderman @looney2ns @bigredfish @aristobrat and many members here, they always do fantastic for helping all here.
 

Parley

Known around here
Joined
Dec 19, 2015
Messages
5,604
Reaction score
15,939
Location
Cypress, California
My final summary, thoughts recommendations added into post #5
From your post #5. "At this point I think most of us are aware of the performance of the 1/1.8” sensors that Dahua is using and the really remarkable image that these are capable of in various cams". I could not agree more. I think Dahua has found the sweet spot with these cameras and the B5442E series cameras. As I have stated in another post they are outperforming my 2MP Hikvision cameras with their 1/1.8" sensor and I paid 2 1/2 times more money for them. I will be placing and order next month for two more in the B5442E series and that will give me 10. :) The only reason I am waiting until next month is for budgetary reasons. I have a self imposed spending limit. ;)
 
Joined
Jul 31, 2018
Messages
1,306
Reaction score
2,468
Location
texas
From your post #5. "At this point I think most of us are aware of the performance of the 1/1.8” sensors that Dahua is using and the really remarkable image that these are capable of in various cams". I could not agree more. I think Dahua has found the sweet spot with these cameras and the B5442E series cameras. As I have stated in another post they are outperforming my 2MP Hikvision cameras with their 1/1.8" sensor and I paid 2 1/2 times more money for them. I will be placing and order next month for two more in the B5442E series and that will give me 10. :) The only reason I am waiting until next month is for budgetary reasons. I have a self imposed spending limit. ;)
I have a significant other imposed spending limit. But she will get over it because i have 5 more cameras to get for my rebuild system and that includes this one, i might need two, but we will see.
 
Top