Review: EmpireTech IPC-T58IR-ZE-S3 / Dahua 5842T-ZE-S3 - 8MP / 4K Turret Camera

Another note (and plus point) about the S3's for those that weren't aware is that when you choose the Video Metadata function (AcuPick or Standalone) they do show up with ANPR functionality. Now, with that said, this is not implemented well yet and a bug I've been speaking with Dahua about BUT critically it IS supported by these cams. The Dahua NVR's will pickup on this attribute as well. However current (including the latest September) FW has not enabled the functionality correctly. I'm hoping Dahua will stand behind this and correctly implement soon. As I said, I've provided a ton of information and recommendations on this, it's now on them.

Can You tell more about disabled ANPR on S3 cams?
With AcuPick enabled I don't see even plate snapshots.. Only in VMD standalone..
But not plate OCR at all on 5xx2-S3...
 
I hope that next gen of cams & NVR will solve that.
Running IVS + AcuPick 1.0 (3d modeling) + Video Metadata with attributes + Face Detect/Recognition + ANPR on one cam should be simply functionality available without any limitations on PRO camera / NVR.

A few weeks ago HIK presented new generation of DeepinMind cams (aka DarkFighter 2.0)..
They allowed to run 2 AI algorithms together in most combinations (like IVS + VMD on cam).
And added AcuSearch 1.0 as part of IVS.

But there are still some limitations (some functions like face mosaic or ShotN disables second AI function)..
And ANPR is not a part of Video Meta Data but separate mode...
 
Yes, I have a few 7442-Z4-X in different locations...
Works very well for ANPR duties in VMD mode (in color at day, in IR mode at night with very well visible cars)..

Yes these are great cams, always enjoy them BUT I don't like the mounting options they went with on design. Cameras themselves are fantastic BUT if they built them to interface with regular PFA square mounts OR even had mounting kits to allow mounting into those, would be ideal.

5xx2-S3 is doing that.. sending plate snapshot but there is no plate OCR..
yes, they don't do ANPR on 5xxx-S3 to not cannibalize 7442...
But also due more limited SOC power in 5xx2-S3...

Yes there are limits on the SOC's in the 5x S3's BUT they are not being pushed yet. The S3's represent a series of cams where for the first time in recent time (for 5x series) they are not hitting the theoretical or real world limits of the SOC, yet.

Also 7442-X don't have normal AI functionalities like AI-SSA or AI-Codec.
This is done to spare one of the NPUs (AI processors) in HiSilicon SOC in 7442-X.

HiSilicon SOCs usually have two almost identical NPUs processor to run 2 different AI models in parallel.
In 5xx2-S3 one is used to run AI-SSA / AI-Codec models and second one to run selected in Smart Plan AI algorithm.

in 7442-X two NPUs are free to run selected in Smart Plan functions - it can run 2 different smaller algorithms (like Face Detect + Face Recognition) or one bigger like ANPR or Video Meta Data (probably OCR part is run on second NPU core)..

Yes the NPU usage in 7x series is different and certainly more methodical in execution. My hope is that with NPU functionality (and price) dropping that an S4 series of cams will truly knock the capabilities and critically, real world performance, out of the park.

1. there are compromises - there are lots of bugs in firmware & implementation. Of course every generation of firmware or every generation of cams fixes some, but also adding a new ones...

2. Actual generation of cams (5xx2-S3, 7xx2-X) and NVR (5xxx-EI, 6xx-XI) added possibility to run more that one AI algorithm - which gave us much more functionality outside simple IVS. But there are still huge limitations which functions / algorithm work together with which one in what configuration or what functionality we are losing when enabling another one.

I hope that next gen of cams & NVR will solve that.
Running IVS + AcuPick 1.0 (3d modeling) + Video Metadata with attributes + Face Detect/Recognition + ANPR on one cam should be simply functionality available without any limitations on PRO camera / NVR.

Yes I believe we're getting there. The Xi NVR's for example are incredibly good at what they do. The S3 cams are also very capable as I've mentioned and shown here. Some of the limitation are self imposed by Dahua and others (hence what I mentioned above), others are regressions which in any dev work, can exist BUT critically need to be resolved. This is one of the things I try to work on with Dahua in every release and in an ongoing fashion because TBH there are a lot of regressions and a lot of issues. I'm pleased to say that this work (although takes me many hours) has led to some great wins for all of us. The sad part is that for Dah, Hik and others, the lifecycle of the cams / NVR's now is so short. All are very quick to move onto new products and EOL the old. In a lot of cases as we both know, this means new SOC's with new functional code which means that last generation (and certainly 2 or more generations back) fail to ever get the needed fixes (regression, feature or tweak), in place due to dev priority being focused elsewhere. Thats why what we do here in this industry is so important. Raising awareness to manufacturers on critical need, focused issues etc in efforts to get a better (or improved) end product. Similarly though its just as critical in educating end users/business on how to scope, install, configure and deploy correctly and in an optimized fashion for each location, required target capture environment, threat model (where appropriate), as part of an overall solution.
 
  • Like
Reactions: steve1225
Can You tell more about disabled ANPR on S3 cams?
With AcuPick enabled I don't see even plate snapshots.. Only in VMD standalone..
But not plate OCR at all on 5xx2-S3...
  1. Yes, they do support it just not implemented / purposefully disabled in current FW. Couple of ways you can see this:
    1. Throw on AcuPick as an example, go into your NVR of choice (I currently have these hooked into both EI's and XI's) go into AI Settings (on EI) or AI / Parameters on XI.
    2. Then go down to ANPR, choose the channel you have the SR camera(s) on and you will see that you can enable it for the cam (not to be confused with NVR AI processing), this is strictly reading the supported attributes of the cam itself. You won't get the 'Config is Not Supported' error that you would with a truly unsupported cam (think S1 or S2 versions that do NOT support at all).
    3. You'll notice (depending on NVR) you can enable Block List Sync etc (XI where you can set database local or remote).
  2. On the cam you can sanity check as well. Go into Picture (for captured snapshots) in the camera and you'll see the drop-down of capture types.
    1. Choose ANPR and you'll see it captures 2 (of the usual 3) pictures, wide car, close up but critically even though the camera knows it can capture ANPR, does NOT, currently show Plate due to OCR being disabled in current FW.
    2. Critically IF you go into the Record menu (for video capture) and then choose ANPR from the selection, you will see NO video is captured at all, even when the Picture / Snapshots ARE recognizing and tagging the vehicle correctly as an ANPR capture.
An additional data point that I find very interesting in my testing is that IF you have an NVR with ANPR processing via AI on it (like XI), the exact grab of the plate (again processed on NVR) matches the exact snapshot (even by time) that the S3 cam is processing for ANPR in camera (snapshot), like I said difference is that cam doesn't complete the output).

This is VERY similar to what Dahua did with the early 49xxx that I mentioned as well except it was the reverse. In that implementation of FW for the S1 series cams they DID full ANPR up front (but was not called out as a feature) as part of Video Metadata processing and tagging. They then, under the covers unfortunately released new FW for S1 that disabled it (only showed car wide and close video + pics) and was missing completed in the S2 series of the same cam.

In the case of the S3, this remaining 10% functionality (the actual plate processing) appears to be completed but output and associated tagging is disabled at the moment. Again Video doesn't tag ANPR at all (currently) but Snapshots DO up to the point of the final display/using OCR output. It is 100% currently disabled in current FW. However, it is also 100% possible to deliver it, since the camera a) has the associated attributes, tagged and b) is only limited but the current FW release. My hope is that with what I've shown Dahua and how code is correctly attempting to process, that they do unlock that gate.
 
Yes these are great cams, always enjoy them BUT I don't like the mounting options they went with on design. Cameras themselves are fantastic BUT if they built them to interface with regular PFA square mounts OR even had mounting kits to allow mounting into those, would be ideal.

OMG... I hate mounting system & junction box on 5xx2H/7xx2..
Usually You can loosen the screw to correct the camera position only 2-3 times..
After that the screw is good for nothing.

Also there is no space for all 10? cables with plugs going outside cam in included junction box..
And those cams are not compatible with any other on the market..


Yes there are limits on the SOC's in the 5x S3's BUT they are not being pushed yet. The S3's represent a series of cams where for the first time in recent time (for 5x series) they are not hitting the theoretical or real world limits of the SOC, yet.

I'm not sure - in first version of internal Dahua presentation about AcuPick 2.0, there were presented more human/car attributes for 5xx2-S3..
they cut the list in second version & final firmware..

Maybe CPU/GPU part of 5xx2-S3 SOC didn't hit limits, but it looks like NPU part had..

Yes the NPU usage in 7x series is different and certainly more methodical in execution. My hope is that with NPU functionality (and price) dropping that an S4 series of cams will truly knock the capabilities and critically, real world performance, out of the park.

HiSilicon is main SOC manufacturer for all Chinese mobiles..
SOC for mobiles are the fastest growing category of integrated circuits, which is pushing the entire SOCs production market forward.
SOCs for CCTV are simply cut / limited / cheaper versions of SOCs for mobiles.

Yes I believe we're getting there. The Xi NVR's for example are incredibly good at what they do. The S3 cams are also very capable as I've mentioned and shown here. Some of the limitation are self imposed by Dahua and others (hence what I mentioned above), others are regressions which in any dev work, can exist BUT critically need to be resolved. This is one of the things I try to work on with Dahua in every release and in an ongoing fashion because TBH there are a lot of regressions and a lot of issues. I'm pleased to say that this work (although takes me many hours) has led to some great wins for all of us. The sad part is that for Dah, Hik and others, the lifecycle of the cams / NVR's now is so short. All are very quick to move onto new products and EOL the old. In a lot of cases as we both know, this means new SOC's with new functional code which means that last generation (and certainly 2 or more generations back) fail to ever get the needed fixes (regression, feature or tweak), in place due to dev priority being focused elsewhere. Thats why what we do here in this industry is so important. Raising awareness to manufacturers on critical need, focused issues etc in efforts to get a better (or improved) end product. Similarly though its just as critical in educating end users/business on how to scope, install, configure and deploy correctly and in an optimized fashion for each location, required target capture environment, threat model (where appropriate), as part of an overall solution.

Amen, brother :)
Every time I deal with Dahua Poland representatives, I literally go on a tirade about the stability and quality of their software. Especially mobile applications (DMSS) and for computers. But also firmwares.

I am a programmer and I can determine what is wrong with their apps. Unfortunately, few people want to hear that..
From what I have noticed, people at Dahua Poland are afraid to report problems in software or firmware to Chinese HQ.
And they are even more afraid to enforce so that the Chinese will fix it.

The only situation where they doing more are when there is huge project (hundreds / thousands of higher models cams) which is at risk (its sale) due to some errors or lack of functionality.
 
The only situation where they doing more are when there is huge project (hundreds / thousands of higher models cams) which is at risk (its sale) due to some errors or lack of functionality.
I did software for a large tech company, and could have written the same thing. One of the workers' mottoes was "There's always time to do it over, but never time to do it right".
 
  • Like
Reactions: bigredfish
You can try 7442-X..
It have full ANPR in Video MetaData mode (plates are OCR & searchable)..
Also have different video processing - I found that on 7442 I can setup low gain (like 30 max 40 for low noise) with short shutter for night mode & then do SSA/WDR which on 7442 will do second level gain, much better that 5442. 7442 works much better for motion at night conditions.

7442 (but only 4mpx version) have also time division exposure technology - which joins static/slow moving and fast moving objects from different shutter exposures.. something like WDR/HDR for motion...



Also wait for AcuPick 2.0 firmware for 6xx-XI...
Dahua promised that 6xx-XI will have option to do AcuPick on recorder & Face Detection/Recognition for AcuPick cams..
Release date for 6xx-XI promised by Dahua in AcuPick 2.0 presentation was July 2024.
But this didn't materialized.

ps. you can use HDMI over CAT5 extenders with KVM functions (transfer USB mouse back to transceiver)...
Thanks, unfortunately its hard to get the customer to spend on the 7xx models. Hoping they may come out with cheaper APNR models in the future.
Definately waiting on Acupick 2.0. Thought we would be able to use AcuPick 1.0 over web service with this XI but guess we have to wait for now.

Also separately on the 5xx EI when you add Face Detect by Recorder you only get 4 channels of SMD by Recorder.
Removing Face Detect you get full 8 channels of SMD by recorder.
I also found SMD by recorder doesnt label the Human/Vehicle events properly alot of the time.
Most new cameras I am using have Human/Vehicle detect now though.

Lots of good info in this thread, will spend some time reading the latest posts.
 
Thanks, unfortunately its hard to get the customer to spend on the 7xx models. Hoping they may come out with cheaper APNR models in the future.
Definately waiting on Acupick 2.0. Thought we would be able to use AcuPick 1.0 over web service with this XI but guess we have to wait for now.

Also separately on the 5xx EI when you add Face Detect by Recorder you only get 4 channels of SMD by Recorder.
Removing Face Detect you get full 8 channels of SMD by recorder.
I also found SMD by recorder doesnt label the Human/Vehicle events properly alot of the time.
Most new cameras I am using have Human/Vehicle detect now though.

Lots of good info in this thread, will spend some time reading the latest posts.

You can use AcuPick 2.0 features without AcuPick 2.0 interface on 600-xi..
setup AcuPick on camera..
on recorder in AI search -> human / vehicle menus You will have possibility to search human / vehicles by attributes like in AI Search -> AcuPick menu on 5xxx-EI..
in reality Acupick 2.0 = AcuPick 1.0 (3d modeling, searching by image) + simplified Video MetaData (searching human / vehicles by attributes)..

5xxx-EI have 8 AI units.. SMD and Face Recognition (when face detection is done on cam) is using 1 AI unit, IVS 2 AI units, Face detection & Recognition 4 AI units...

Why are You using SMD? it creates a lot of false positives even when run on camera..

In AcuPick 2.0 mode on 5xxx-S3 cams, You have both: mini IVS (tripwire/intrusion) and mini Video MetaData (human/vehicles without face/plate snapshots)..
for perimeter protection use IVS..

If You want faces, better solution is disable AcuPick on camera and use full Video MetaData... it generates human/vehicles/faces/non motor vehicles records with faces and plates snapshots and it works very well.. I run all external / street / gate cams this way on all installations with a new 5442-s3..

if you will contact me by DM, I can send you demo pages from searches on NVR done from 5442 or 7442 with full Video Metadata (no AcuPick mode)..
it can generate thousands records per day depending of street traffic.
 
  • Like
Reactions: bigredfish
You can use AcuPick 2.0 features without AcuPick 2.0 interface on 600-xi..
setup AcuPick on camera..
on recorder in AI search -> human / vehicle menus You will have possibility to search human / vehicles by attributes like in AI Search -> AcuPick menu on 5xxx-EI..
in reality Acupick 2.0 = AcuPick 1.0 (3d modeling, searching by image) + simplified Video MetaData (searching human / vehicles by attributes)..

5xxx-EI have 8 AI units.. SMD and Face Recognition (when face detection is done on cam) is using 1 AI unit, IVS 2 AI units, Face detection & Recognition 4 AI units...

Why are You using SMD? it creates a lot of false positives even when run on camera..

In AcuPick 2.0 mode on 5xxx-S3 cams, You have both: mini IVS (tripwire/intrusion) and mini Video MetaData (human/vehicles without face/plate snapshots)..
for perimeter protection use IVS..

If You want faces, better solution is disable AcuPick on camera and use full Video MetaData... it generates human/vehicles/faces/non motor vehicles records with faces and plates snapshots and it works very well.. I run all external / street / gate cams this way on all installations with a new 5442-s3..

if you will contact me by DM, I can send you demo pages from searches on NVR done from 5442 or 7442 with full Video Metadata (no AcuPick mode)..
it can generate thousands records per day depending of street traffic.
Do you only use motion detection + ivs and metadata?
 
Do you only use motion detection + ivs and metadata?
Depends on your use case. You can just use full blown Video Metadata by itself in non-Acupick mode + Motion Detection IF you really want it. Personally I would stay clear of MD + SMD and instead as you know, I advocate for Video Metadata or now, AcuPick.
 
  • Like
Reactions: bigredfish
Thanks, unfortunately its hard to get the customer to spend on the 7xx models. Hoping they may come out with cheaper APNR models in the future.
Definately waiting on Acupick 2.0. Thought we would be able to use AcuPick 1.0 over web service with this XI but guess we have to wait for now.

Also separately on the 5xx EI when you add Face Detect by Recorder you only get 4 channels of SMD by Recorder.
Removing Face Detect you get full 8 channels of SMD by recorder.
I also found SMD by recorder doesnt label the Human/Vehicle events properly alot of the time.
Most new cameras I am using have Human/Vehicle detect now though.

Lots of good info in this thread, will spend some time reading the latest posts.
XI's are really good, I have one and love it BUT it is currently trailing behind the EI from an AcuPick 2.0 perspective. I have however addressed this with Dahua and asked that they release a new FW version ASAP for XI that looks to get to feature parity with EI, especially from an AcuPick perspective. This is needed, especially when Dahua themselves call out the XI in the AcuPick 2.0 press. XI is phenomenal WRT on NVR processing, especially from AI by recorder ANPR. EI is surprisingly good for the market its aimed at for sure and certainly a great value for money but it is lacking in certain on NVR AI areas. With that said, AcuPick 2.0 and specifically Map Tracks is very nice to see.

I plan on doing a deep dive for people here on EI vs XI, AcuPick vs VMD etc but at the moment am focused on an LPR challenge that was set (should have that up in the next day or 2) + the bug documentation for Dahua.
 
  • Like
Reactions: cpd5215
Depends on your use case. You can just use full blown Video Metadata by itself in non-Acupick mode + Motion Detection IF you really want it. Personally I would stay clear of MD + SMD and instead as you know, I advocate for Video Metadata or now, AcuPick.
I currently only have Tioc cameras with Ivs function and use a 5208-p-ei recorder. I won't get my s3 cameras until next month.
 
I currently only have Tioc cameras with Ivs function and use a 5208-p-ei recorder. I won't get my s3 cameras until next month.
Then use IVS as your main. The S3 cameras will blow your mind BUT as I continue to say (and will do in each video I do on these), you MUST dial in these carefully especially depending on target use case
 
Then use IVS as your main. The S3 cameras will blow your mind BUT as I continue to say (and will do in each video I do on these), you MUST dial in these carefully especially depending on target use case
Okay thank you. I bought the IPC-HFW5442H-ZHE-S3 version. Good choice or ?
 
Last edited:
  • Like
Reactions: bigredfish
XI's are really good, I have one and love it BUT it is currently trailing behind the EI from an AcuPick 2.0 perspective. I have however addressed this with Dahua and asked that they release a new FW version ASAP for XI that looks to get to feature parity with EI, especially from an AcuPick perspective. This is needed, especially when Dahua themselves call out the XI in the AcuPick 2.0 press. XI is phenomenal WRT on NVR processing, especially from AI by recorder ANPR. EI is surprisingly good for the market its aimed at for sure and certainly a great value for money but it is lacking in certain on NVR AI areas. With that said, AcuPick 2.0 and specifically Map Tracks is very nice to see.

I plan on doing a deep dive for people here on EI vs XI, AcuPick vs VMD etc but at the moment am focused on an LPR challenge that was set (should have that up in the next day or 2) + the bug documentation for Dahua.

You can send them this slide - this is Dahua AcuPick 2.0 presentation from June, where they promised AcuPick 2.0 firmware for 6xx-XI in July...

Screenshot 2024-06-12 at 14.08.02.png
 
  • Like
Reactions: Timokreon
Okay thank you. I bought the IPC-HFW5442H-ZHE-S3 version. Good choice or ?
Again, depends on what you want to use it for target wise (including distance to, type of, lighting etc) but for general surveillance then yes thats a good model. Consider the Z4 version if you want something with extra reach. That is the one I am doing the challenge with currently.
 
Do you only use motion detection + ivs and metadata?

If You enable FULL Video MetaData (without AcuPick) then You have only VMD on that camera. there is no SMD or IVS - full AI power is going to FULL version of SMD with all features (faces, car plates snapshots etc).. You can add IVS on recorder (AI by recorder) for that camera...

I use FULL VMD on street cams - so usually don't need IVS.. VMD gives me much more..

No one here is using SMD at all - SMD is a toy..
 
  • Like
Reactions: bigredfish
If You enable FULL Video MetaData (without AcuPick) then You have only VMD on that camera. there is no SMD or IVS - full AI power is going to FULL version of SMD with all features (faces, car plates snapshots etc).. You can add IVS on recorder (AI by recorder) for that camera...

I use FULL VMD on street cams - so usually don't need IVS.. VMD gives me much more..

No one here is using SMD at all - SMD is a toy..
Okay, I see. I'll use it when the two cameras arrive.