IP Cam Talk

Welcome to the new IPCT! If you are having an issue logging in, please clear your cookies / cache.

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

Wildcat_1

Getting comfortable
Joined
Dec 9, 2018
Messages
260
Reaction score
350
Location
US
All, thanks to Andy (@EMPIRETECANDY ) we have a new 4MP Ultra AI camera for testing. This camera has been provided by Andy in exchange for a fair and honest review.

This is a varifocal bullet cam with a 2.7mm to 12mm optical zoom and Ultra AI feature set. Full specs below and will be adding initial pics & vids in post 2, AI feature breakdown and testing in post 3, tips and tricks on how to make these features work and other settings / menu functions in post 4 and a summary of findings / recommendations in post 5. Trying this approach to the layout to make it easier to find for everyone going forward.

As there will be a ton of features to share with all of you as well as the standard image quality tests, once those are completed I may run comparison vids/pics (if interested) against some of the cams below with the B5442 being the obvious choice for comparison due to sensor performance.

  • 5231 Pro (1/2.8” F1.6 2MP HD sensor)
  • 5241 Pro AI (1/2.8” F1.5 2MP sensor)
  • B5442 Pro AI (1/1.8” F1.8 4MP sensor)
  • 2831 Lite (1/1.8” F1.9 8MP 4K sensor)
  • 5831 Pro (1/2.5” F1.4 8MP 4K sensor)

I plan to review in multiple lighting conditions including extreme low light to test the Starlight+ features. Also intend to list some of the unique features of this particular camera not limited to its Ultra AI functionality but also FW / menu differences as well.

You will notice that in testing the cams I like to show not only the (in your face) quality tests but also some real world install tests in both close up and overview locations to push the camera and sensor as much as possible.

If there is anything anyone wants to see specifically, please let me know and will try to accommodate. Also be sure to check out @looney2ns review of the 8MP (4K) version of this cam here


Price from @Empirectecandy is approximately $480 and can be purchased here:
Aliexpress link - US $480.0 5% OFF|New Model IPC HFW7442H Z 4MP AI IR Bullet Network Camera, free DHL shipping-in Surveillance Cameras from Security & Protection on AliExpress - 11.11_Double 11_Singles' Day


Now onto the specs


Image Sensor - 1/1.8” 4Megapixel progressive CMOS - 2688 (H) × 1520 (V)​
ROM - 4 GB​
RAM - 2 GB​
Auto/Manual - 1/3 s–1/100000 s​
Min. Illumination - 0.001 Lux@F1.2 (Color, 1/3 s, 30 IRE)​
IR Distance - 50 m (164.04 ft)​
IR LEDs Number - 4 (IR LEDs)​
Focal Length - 2.7–12 mm​
Max. Aperture - F1.2​
Field of View
Horizontal:111° (W)–47° (T)​
Vertical: 59° (W)–26° (T)​
Diagonal: 135° (W)–51° (T)​
Close Focus Distance
W: 1 m (3.29ft)​
T: 2.5 m (8.20ft)​

AI Features

Perimeter Protection
Tripwire, intrusion (Recognition of vehicle and people), fast moving, loitering detection, parking detection and people gathering.​
Face Detection
Face detection, track, priority, snapshot, face enhancement, face exposure, face attribute extract, 6 attributes and 8 expressions: Gender, age, glasses, expressions (anger, sadness, disgust, fear, surprise, calm, happiness and confusion), mask, beard Face matting setting: face, single inch photo Three snapshot methods: real-time snapshot, priority snapshot, quality priority​
ANPR
ANPR, track, priority, snapshot​
Vehicle attributes: license plate, vehicle type, vehicle color,vehicle brand​
Other attributes: sun visor, seat belt, smoking, phoning, in-car accessories, annual vehicle inspection mark​
Video Structurization
Vehicle detection, non-motor vehicle detection, personnel detection, face detection, track, snapshot, optimization, report high quality picture, vehicle attributes (license plate, vehicle type, vehicle color, and vehicle brand), other attributes (sun visor, seatbelt, smoking, phoning, in-car accessories, and annual vehicle inspection mark), non-motor vehicle attributes: vehicle type, vehicle color,driver number, cloth type, cloth color, and hat), personnel attribute (cloth type, trousers type, cloth color, trousers color, knapsack, and gender), face attribute (gender, age , glasses, mask, and beard)​
People Counting
People counting (count the entry number, exit number and pass number), display and output daily/monthly/annual statistics, No. in area​


Pics Of The Cam

BTW please excuse the install pics. As you can see I usually like clean installs in PVC or conduit but in the case of the test have to string a wire rather than make it look complete :)


Side View Of The Cam

7442 Side Image.jpg


Back View of The Cam
Showing Built In Connections For Junction Box - YES THATS A 3.5MM MIC INPUT, WAHOO !!!! I will be adapting one of my pro shotgun mics for further testing here on sound quality.

7442 Back of Camera.jpg


1st Install Location Pic
Showing Built In Junction Box Mounted Flush Horizontally

1st Location Install Pic 1.jpg


1st Install Location Pic 2
Showing 7442 (Bottom) Next to A 2831T

1st Location Install Pic 2.jpg


2nd Install Location
Next to a 5231-Z12 Showing Built In Junction Box Mounted Flush Vertically

2nd Install Location Pic.jpg
 

Wildcat_1

Getting comfortable
Joined
Dec 9, 2018
Messages
260
Reaction score
350
Location
US
Post # 2 - Reserved For Initial Pics & Vids

First Video - Initial Daytime Video + Associated AI Captures

This first daytime one focuses on a couple of real world install scenarios (overview, close up porch/entry) and I decided to edit together the vid caps with their AI pic captures. I’m using Video Structuralization as the Smart Plan and will go into more detail in the AI post over the coming days with how best to set this up (there are some tips and tricks for better AI success), ANPR and People captures along with differences between what you can accomplish in cam vs AI NVR.

Without going into too much detail in this post, Video Structuralization (VS) is really the Smart Plan you want to use with these AI cams. Sure you can setup ANPR or People detection separately but VS allows you to capture both without having to give up one PLUS also enables Face Detection as well. Therefore you are getting 3 AI features using 1 Smart Plan.

Will post Night next so you can see how this performs there as well. In the AI post I also put EVERY attribute to the test, if the cam says it can detect it, I try it :). Enjoy !

Remember to change your Youtube settings.



Initial Night Video


All, here are some initial night video’s. Settings as stated and I purposefully did not dial the image in like I would normally as I wanted to go with bare minimum changes in these first videos. I neglected to say in the first video that lighting is 1 x 100w bulb on the porch, the side shots across the stop sign are lit by 1 LED street lamp (will grab an iPhone pic to show what it can see later). Will be completing more night time testing as well.

Also, as an ANPR / LPR individual (run 5231-Z12’s and 5241-Z12) I will be the first to tell you that the angle and height I mounted the cam at for this initial test is ABSOLUTELY NOT the way to position an ANPR / LPR BUT I wanted to show what is possible with this cam even in crazy, real life locations (punchline, I’m impressed) :)

Unfortunately as Youtube compresses the crap out of everything the videos usually take a bit of a beating so I built this from a ProRes proxy and uploaded that way which allowed a little more retention of detail etc

More to come in the AI post and I will be testing more locations, showing whats captured by the NVR in cases of the ANPR / Person Detection and what can be achieved when dialing in the image. Enjoy !


 
Last edited:

Wildcat_1

Getting comfortable
Joined
Dec 9, 2018
Messages
260
Reaction score
350
Location
US
Post # 3 - Reserved For AI Feature Testing & Results

3.1 AI FEATURES - INITIAL ANPR

Now I couldn’t let all of you go into the weekend without giving you a taste of what everyone’s been waiting for……ANPR / LPR. Now as I mentioned, the first test (this one) was to really push the camera to the test, I would never setup an LPR camera on purpose at this angle or height but wanted to see what it could do in a regular overview position.

The way in which this camera works and what the end result is depends on where you are sending the feed.

On Cam (SD Card recording)

If you are only using on cam (SD card) then you will get :
  • Separated video and pictures that are tagged CarDetect, assuming you are using Vehicle Detection under ANPR rule or my favorite Video Structuralization.
  • As separate files do not link to each other when recorded in cam and would have to be viewed separately.
  • Video does not have any overlay of the details of the capture, yet it does show on the web gui if watching live (the clips that pop up) and of course is burned into the picture.
You can configure the Vehicle Detection capture (in Video Structuralization) a number of ways but I configure so if it detects a plate then I end up with 3 pictures (full res wide shot with car details, plate close up and close view (vehicle and plate).

AI NVR

When using an AI NVR you get:
  • Pictures and video’s together under Vehicle Detection in its AI menu
  • Tags all plate info so you can search by it later, very cool !
  • Due to last bullet item you can then search using the AI function for attributes such as Plate No., Color, Type, Logo, Plate Color, Ornament, Calling, Seatbelt.
  • If you don't want to search by attributes then leave this completely open (ALL) and when you hit search it will pull back pic and video matches together !
  • In conjunction with an AI NVR it also allows you to build Black & White Lists for plates
--
As video in cam does not overlay, lets take a first look at ANPR/LPR pics and cover AI later.

So, in this first example (with bad angle and height) at full zoom, how does the camera do…….in my opinion its VERY impressive but let me know your thoughts. I’ve obscured partial plates and faces to protect privacy. When you look at what its trying to capture from a PPI perspective, its amazing.

If you have the camera web gui up and Video Structuralization set as on-screen you get to see the results in real-time which also (in the case of the Ford) showed on screen that they were wearing seatbelts (unfortunately didn’t capture the screen grab in time).

DON’T WORRY MORE TO COME FOR ANPR/LPR AS WELL AS THE OTHER AI FEATURES

Couple of pics of vehicle capture
Each captured the plate accurately and when you look at the distance to target (70-80ft) its doing a great job based on a 2.7 to 12mm lens especially since the DORI rates this as just under 42ft (at full zoom like this) to identify and that would be a person not a plate :)


Black Toyota.jpgBlue Subaru 1.jpgBlue Subaru 2 - Plate.jpgBlue Subaru 3 - Close.jpgBlue Taurus 1.jpgBlue Taurus 3 - Plate.jpgBlue Taurus 2 - Close.jpgWhite 4 Runner 3 - Wide.jpgWhite 4runner 4 - Plate 2.jpgWhite 4 Runner Capture 1.jpgWhite Mazda Capture 1.jpgWhite Mazda Capture 2 - Plate.jpgWhite Pickup 1.jpgWhite Pickup 2 - Plate.jpgWhite Toyota 1.jpgWhite Toyota 2 - Plate.jpgWhite Toyota 3 - Wide.jpg


Video Structuralization Meta Clip From Web GUI

Video Struralization - Capture INterface 1.jpg


3.3 Night ANPR / LPR - Coming SOON

NEW 11/4


Ok, I know everyone wanted to see ANPR / LPR night captures + low light. Well, here is both in the same cap :). Let me first call out some info:
  • Snow was still an issue but made it work :)
  • Cam at full zoom @ 1/500
  • Sleet coming down with splash back from road made for tricky captures
  • Still not ideal angle or height compared to my 5231 & 5241 LPR setups BUT the best I can do with this particular focal length for now as I wanted to show real-world installations rather than a simulation
  • Camera moved to other location for ANPR / LPR which is completely dark except for some IR flashback due to the location
  • NR turned off which introduced noise but with the sleet I wanted to try and combat any further loss in detail caused by NR
  • EVERY vehicle captured here resulted in a full ANPR hit
  • Editing was to obscure the plates for privacy
  • The ANPR defaulted to a lot of info when this dark (such as color etc) although plate info was accurate
  • Mix of clean and dirty plates to show what the camera is still able to do (check the E36 for example of a poor / tough situation where the cam still got a usable capture)
  • Included slo-mo’s so you can look at each live frame then the associated capture
  • Cam is applying its own processing (sharpening + enhancement) automatically to final image which although you don’t control (like you can in Face Detection), does a good job of pulling the plate out.

Enjoy !


----

As the snow kept coming down and I waited for the round 2 of the storm, I knew I couldn’t move the camera so instead got some further night shots which again, even at the bad height / angle accurately hit all ANPRs on the vehicles in these clips, EVEN WHEN IT WAS SNOWING ! It really does do an amazing job even in poor conditions.

I even gave you all some slo-mo video so you can go frame by frame. The original (non compressed for youtube) you can see much greater detail. I will be moving the camera to a more optimal, in your face location as soon as the temperature improves above 6 fahrenheit / -14 celsius.




3.4 People Detection - Attribute Testing

As I wait for the snow storm to stop so I can test more of the Night ANPR / LPR, I thought I would show you one of the other areas I committed to, testing this cam against all its stated People Attributes. Now as this was a warmer, sunnier day you can imagine what this must have looked like for others as they walked by BUT hey, wanted to test the cam. I also have another interesting twist on this test coming up which I will post as soon as I can.

So here’s how it did. It takes 2 pics under Human Detection AND if it can detect it (and you have enabled) can also grab Face Detection on top. This is why I love using Video Structuralization as it allows you to stack many AI features (definitely consider this tip !). In most case the AI did a great job except it could not detect the backpack, everything else worked.


P8 Umbrella 1.jpgP8 Umbrella 2.jpgP8f-1.jpgP8f-2.jpgP9-1.jpgP9-2.jpg
 
Last edited:

Wildcat_1

Getting comfortable
Joined
Dec 9, 2018
Messages
260
Reaction score
350
Location
US
Post # 4 - Reserved For Tips & Tricks, Dialing In The Cam
I mentioned I would share some tips as we go through so here is 1 that I wanted to share that made a big difference to ANPR capture for me. Hope these help !

Tip #1 - Configuration of Triggers For Video Structuralization Smart Plans - ANPR

Most here are used to setting up Detect Regions in IVS and normally its just the area you want to look at then your choice of recording that area (intrusion zoning) or when a specific tripwire is triggered. What I found in testing the Ultra AI series is there are some differences in the way this operates when using for things like ANPR. Specifically:
  • Detection region triggers quite quickly (not normally a bad thing) which can cause the ANPR processing to kick in too early.
  • The Ultra AI has an ‘Optimized Snap’ mode which although should use the camera AI to decide the best time to take the picture, in tougher or more tight environments (limited window for capture, angles etc) this often didn’t lead to the best results for final capture.
  • The AI processing that delivers the ANPR capture details (metadata burned into the screen, shown in Video Structuralization Live View and sent to AI NVR's as well, if using them) takes place between the Detect Region & the Snap area (Optimized or Tripwire) and is automatic.
The above can lead to non optimal captures or missed ANPR opportunities. The solution is actually two fold:
  1. If you look at the image below you can see that rather than Optimized Snap, I used Tripwire and on screen you can see where I drew that (diagonally across the stop sign). This gave much better results of plates than Optimized in this setup.
  2. You’ll see the Detect Region on the road to the left of the image, this was before I adjusted it and after this screen cap I moved it to the right about halfway up that street (closer to the stop sign). This meant that the camera didn’t try and process too early, captured the vehicles in the sweet spot for this install location AND gave a narrower window for the AI cap and associated stills.

AI Video Structuralization Screen Grab.jpg
 
Last edited:

Wildcat_1

Getting comfortable
Joined
Dec 9, 2018
Messages
260
Reaction score
350
Location
US
Post # 5 - Reserved For Summary / Recommendations
 

Arjun

Getting comfortable
Joined
Feb 26, 2017
Messages
1,204
Reaction score
402
Location
USA
Added to the AI Series Thread,
 

morpheus

Getting the hang of it
Joined
Nov 9, 2015
Messages
128
Reaction score
67
Hello,

I am very interested in the Ultra-AI series and look forward to your review.

Could you answer a few questions?
According to the technical data, the Ultra-AI series can have four substreams instead of two.
1. can you say for which substreams mjpeg is possible as codec?

2. for which substreams is an image section available? (see picture)
Dahua.png

3. can the image be rotated 90° at a resolution of 2688x1520 (full 4MP)? According to the technical data this is not possible, but I can't believe it. (see picture).
dahua2.png


Many thanks in advance and many greetings
Morpheus
 

Wildcat_1

Getting comfortable
Joined
Dec 9, 2018
Messages
260
Reaction score
350
Location
US
@morpheus

Answers to your questions

Q1 - Which sub streams (out of the 4) is MJPEG possible ?​
A - 1, Sub stream 1 allows for MJPEG only​
Q2 - Which sub streams is an image section available ?​
A - Sub stream 2 allows this​
Q3 - Can the image be rotated at 90 degrees @ 2688 x 1520 ?​
A - No, its not possible, will return an error of ‘Current resolution does not support this function), same for 270 degrees. 180 degrees however is allowed. If you want to flip to 90 degrees you have to drop to 1920 x 1080 (HD)​
 

Arjun

Getting comfortable
Joined
Feb 26, 2017
Messages
1,204
Reaction score
402
Location
USA
@Wildcat_1 How do you like the mounting system on this camera? Some users have some reservations over this
 

Wildcat_1

Getting comfortable
Joined
Dec 9, 2018
Messages
260
Reaction score
350
Location
US
Ok everybody, first video is up in Post 2, yes, VIDEO (finally fixed the Youtube issue !!!!!).
 

Wildcat_1

Getting comfortable
Joined
Dec 9, 2018
Messages
260
Reaction score
350
Location
US
@Wildcat_1 How do you like the mounting system on this camera? Some users have some reservations over this
@Arjun the mounting system isn't great but not necessarily due to the shape or design of the outer box but more the cable channel in it. In other words, mounting flush and cabling through the back is relatively straightforward, however when cabling through the main 'conduit' hole that becomes a little trickier again due to the way in which the mounting hole channels the cable.

Therefore due to the nature of the built in junction box there are a couple of things you need to be aware of:

1) The slot in which you feed your CAT cable is very tight. With that said CAT 5 & 6 cables do fit through BUT strain relief boots can cause a problem. Therefore if making the cable once you feed through the slot / channel you should be good, otherwise if you are using a cable you previously made (as I did in this test) then you'll want to pull the strain relief boot back.

2) Due to the almost 90 degree angle of where the cable feeds through the conduit / screw hole (unless installing through the back of the unit), run your cable through the cable slot first before mounting. If you try and pull a fully made (or pre-made) cable (which does fit minus the boot above) it will be much harder due to the aforementioned angle.

Also, be prepared to drill new holes for this mounting unit as the holes don't match previous junction boxes.

Certainly the mounting system is not a deal breaker in my opinion but certainly different and needs a little more thought when installing.

Hope that helps
 

m00st

n3wb
Joined
May 13, 2016
Messages
5
Reaction score
1
Great review. Keen to read how the ANPR works and performs!
 

Davahad

Young grasshopper
Joined
Nov 7, 2016
Messages
64
Reaction score
12
Looking to add another camera to count cars and get LPR from opposite direction. How does this store the count data? Is it just embedded into the video recording?
 

Wildcat_1

Getting comfortable
Joined
Dec 9, 2018
Messages
260
Reaction score
350
Location
US
Looking to add another camera to count cars and get LPR from opposite direction. How does this store the count data? Is it just embedded into the video recording?
Metadata written back to / processed by the backend device i.e in cam or on NVR
 

Davahad

Young grasshopper
Joined
Nov 7, 2016
Messages
64
Reaction score
12
Can you show some reports of what the stats look like for the car counters? Interested in total # of cars for a day and if there is anymore detail (ie., hourly, etc,)
 
Top