Hikvision AcuSense G2 Camera Test

By Rob Kilpatrick, Published Oct 29, 2020, 12:55pm EDT (Research)

********* *** ******** ********* ********** ** ******** ******** *************** "****** **** ** **** *****, animals, ** *** ******** ** *******, for *******, **** ** *********** ***********."

IPVM Image

** ****** *** ****** *** ********* DS-2CD2346G2-I *********:

  • *** **** ******** ** ******* ** Hikvision's ***** ********** ******** *******?
  • *** **** ******** ** ******* ** Avigilon, *****, *** ***** *********?
  • **** ** ****** ****** ****** *** outdoor ******* ** ***** ******?
  • **** ****** *** **** ********** ********?
  • *** ****** ******** ******** *******, *******, or ********?
  • *** **** **** ** ********* **** major *****?
  • **** **** ** *** ** ****?
  • *** **** ***** ******* ******* ** the ***** ********** ********?

Executive *******

** * **** ** *******, ******** G2 ******* **** ***** ***** *******, such ** ***** ****, *******, ******, and *******, **** ******** ********* ** humans (********* ******* *** ********) ** ~10 ***, ****** ********** **** *** performers ** ********* ********* ********. ************, ***** ******* ** *** light *** ****/** ****** ******** **** the ******** ** ******.

*******, ** ******** **** ********** ***** alerts ** *****/***** ******* (*-* *** day) *** *** *********** *** ****** than *** ** ******** ******, **** significant ************ ** ******** ******* ****** backlight.

Non-Production ********

****** *******, ********* ******** ** **** firmware *** *** ******** ********* (**.*.*** build ******) ** ******* ****** ***** alert ******. *******, **** *** ******** introduced ***** ****** ** ****** ******** (shown *****), *** **** ** ********** firmware (**.*.*** ***** ******) — ****** this *** * ******* ***** ** our **** ** ******** **.

** ********* ***** ******* ************ ** firmware ** ******* ***** ******, ** will ****** *** ****** **** ******.

Improvements **** ******** ********

******** ** *** ******* ************ ******** to ** ** *** ******** *******, including ********* ** ******* ********, ****** rejection ** *******, *** ******** ************** of ****** (** ********** ********** ****** as ******** ** ****

  • ********* ** ******* ********:******** ** ********** ****** ******* ********, but ** ********** ******** ******* ******** in *** ******.
  • ****** ********* ** *******:** ***** ******* ** ******* ********** using *** ***** *******, ********** *** target ******** ** "*******" ******** *********** but *** *** ********** *** *** of *** ****** ***** ***** ******** ~1-2 ***** *** ***. ** ******** from ******** ****** ** ******* ********** of ********.
  • ****** *** ********** ** ********:******** ** *** *** *********** ****** as ********. ** ********** ********** ****** as ******** ** **** ******.

*******, ** *** ** ********** ** other ***** ***** ********* ******** ** improved ** ***** *********, **** ** better ***-***** ********* *** ****** ********* area. ** *** ** ********* *********** the **** ** *** ***** *** both ******** ~** *** *** ******** human ******* *********.

Vs. ********, *****, *****

******** ** *********** ** ***** ** that ** *** ********** **** ********* ********* ********:******** ***,***** ***, ******** *********+. ***** *** ** ********* ****** from ********** ****** ** *******, ***** issues **** *** ******** ** *** testing, *** ******* ***** ****** *** be ******* ** **** ************ (*.*., indoors ** **** ***** ** ******** areas).

******** *** ***** ******** ****** *** significantly **** ********* **** ******** **, selling *** ~$*** ** **** **. $200 *** ********. ***** ****** ******* models ***** *** ****** ********* ** Acusense ** *******.

** *** **, ******** ** ************ is ******* ** *** *** *** fixed **** ****** (****** ** ********* models *** *********). ** ***** *******, wider ******* ** *********, **** **** resolutions *** ********* **** *******.

Pricing **. ***-******** ******

*** ********* **-*********-* *** ** ***** online *** ~$*** ***, ******** **** than *** ******** ******** **** (~$*** online).

**** ******* ** * ***** ** moderate ******* **** ***-******** ******. *** example, * *** ********* ***** ****** turret (**-*********-*) ***** *** ~$***-*** *** online, $**-** ***** **** ******** **. Compared ** **** ******* ****** (~$*** ******), ***** ******** **** *********'* **** performing ********* *********, ******* ** **** slightly ******.

"Target ********" *****

******** ** ********** ** ********** "****** Validity" ******* *** ***** ** *** original ******** ****. **** ******* ****** from "*****" ** "*******", ***** ** effectively * ************** ********** *****.

IPVM Image

** *** *******, ** ***** ** drawback ** ***** "*******" ****** ********, with ** ****** ********* ** ******* range ******** ** ***** ********.

False ****** ** ******* ***** ******* ********

******* ******* *** **** ** ***** when ***** "*****" ********. *******, ***** alerts **** ** ****** ******* **** increasing ** "*******."

IPVM Image

False ****** ** ***** / ***** *******

**** ***** ******* ******** ("*****"), *** analytics ********** ******* ** ***** *** large *******. ***** ****** ***** *** be ********** ** ***** ********* ****** Validity ********, **** *-* ****** *** day ***** ********* ** "*******."

IPVM Image

People ********** ******** *******, ********, **********

******** ** ********** ******** ****** *******, crawling, *** ********** **** ** ****** of ****** ** *** ***** ****** testing.

IPVM Image

No ***** ****** ** ***** **** ** *******

***** ****/******* ********, ***** **** *** droplets *** *** ***** *** ***** alerts ** *** *****. ******* *** briefly ** **** *********** **

******* ****** ** *** ***** **** also *** ******* **.

No ***** ****** ** ****** ****** *******

** *** *******, ****** ****** ******* of ***** ****** **** ** ***** light ******* **** *******.

************, ******* ****** ********** *** ***** were **** ******* ** *** ******** G2 ******.

Human *** ******* ********** ****** *********

******** ** **** **** ***** *** vehicle ************** ** ****** *********, *** found ** ***** ********* ******. *******, this ******** *********** ******* ** ********* Detection ***** "*****" ****** ******** ********, resulting ** ***** ****** ** ******* and ******* *******.

IPVM Image

~10 *** ****** *** ****** *********

** *** *****, ******** ** ******** detected ** ~***' ********, ~** *** in *** *** **-*********-* ****** ** tested. ***** *** ************ *** ***, comparable ***** ********** ********* ** *** *********, **** ********** ********** ***.

IPVM Image

Improved **** ***********

**** *********** ******** **** *** **, allowing **** ***** ******* ** ~** PPF.

IPVM Image

Improved *** ***** ***** *******

*** ***** ***** ******* ********, ******** sharper ***** ******* ** ~** ***'

IPVM Image

Slightly ****** **** ***** ***********

** *** *******, **** ***** *********** was ******** ******, ********** ** ***** PPF ****** (~** *** *****), **** slightly ****** ****** ******* *** ****** text **********.

IPVM Image

Poor *** ***********

** *** ***** *** *********** *** worse **** *** ** ******** ******, overexposing *** ******* ******* ** *** scene *** *********** ******** ****** ******* against ****** *********.

IPVM Image

VMS ***********

The ********* ** ****** ******* ********** ****** **** ***********, *********, *** ** ******* *** *** *** ********* **** ******* ** ********.

IPVM Image

Unknown ***

*** ******** ** **** * *** with ** ********, **** *****.

IPVM Image

**** ******* **** ******** ** ***** used ** ********* *** ***** **** a ******** ****** ****** ********** **** (VPU).

Versions ****

*** ********* ******** **** **** ****** testing:

  • **-*********-*:**.*.*** ***** ******

Comments (16)

Does your model have a White LED?

Agree
Disagree
Informative
Unhelpful
Funny

UI1, the DS-2CD2346G2-I that we tested does not have the white light LED.

Agree
Disagree
Informative
Unhelpful
Funny

Rob,

Try to get Gen2 with White LED

It does have an option to turn White LED ON upon Analytics trigger

LED time can be set up to 60 sec (maybe longer)

kinda fun to play

My cam is 8Mp with IR, White LED, and Speaker

Agree
Disagree
Informative
Unhelpful
Funny

What's the model number of that camera and what region are you in?

Agree
Disagree
Informative
Unhelpful
Funny

I am in Canada

HIK OEM

Agree
Disagree
Informative
Unhelpful
Funny

I'm in Canada too. "HIK OEM" doesn't help me. What's the model number? Hik's website doesn't even list their Acusense turrets (let alone any 8MP models) and the only 8MP one at ADI is the DS-2CD2386G2-I with no light/speaker. Only the 4MP DS-2CD2346G2-ISU/SL has those features.

Agree
Disagree
Informative
Unhelpful
Funny: 1

Interesting on the SoC.

While not salient for NDAA, knowing if its a HiSilicon custom SKU for Hik could give insight into potential SoC stock shortfalls in the near term.

Agree: 1
Disagree
Informative
Unhelpful
Funny

Did the beta software eventually get rid of the false positives on animals or did it only introduce false positives on parked cars?

Agree
Disagree
Informative
Unhelpful
Funny

There are two contradictory phrases I found in the article. I just wanted to clarify if the animal false alerts are better with the newer beta firmware.

- "G2 did not false alert on animals when using higher confidence settings, though some issues were still possible using default settings."

- "When using default settings ("Basic"), the analytics frequently alerted on small and large animals. These alerts could not be eliminated by using increased Target Validity settings, with 1-2 alerts per day still occurring at Highest."

Agree
Disagree
Informative
Unhelpful
Funny

The beta firmware lowered the amount of false alarms on animals but still did not get rid of them completely, there were still 1-2 a day using the "Highest" target validity setting. Thanks for pointing that out, I corrected it in the report.

Agree
Disagree
Informative: 1
Unhelpful
Funny

On the SoC, does IPVM think that HIK made that SoC themselves or are they hiding that it is something else?

If HIK made the SoC themselves, would that indicate HIK are potentially very good at designing an SoC internally?

Agree
Disagree
Informative
Unhelpful
Funny: 1

Hikvision has not been claiming to develop SoCs and, in particular, nothing like this for deep learning. The most likely answer is that this is from Huawei, but is unbranded, since Huawei is the largest provider in this space and we have heard previous reports of Huawei shipping unbranded chips in the past to large customers like Hikvision.

Agree
Disagree
Informative
Unhelpful
Funny

When you say most likely Huawei, do you mean Huawei HiSilicon SoC?

While a lot of the smaller (including UNV) Chinese IPC providers are claiming to have changed SoC, Dahua and HIK don't seem to be going the same promoting this. So I suppose I am wondering do they have some secret in-house solution or maybe they have so much stock that they don't need to make rushed changes.

Agree
Disagree
Informative
Unhelpful
Funny

maybe they have so much stock that they don't need to make rushed changes.

See: Huawei HiSilicon Shortage Impacts Surveillance Manufacturers. As Isabella reports, the consensus on the China side is that Dahua and Hikvision are well stocked on Hisilicon chips.

Agree
Disagree
Informative
Unhelpful
Funny

Whatever is happening in the supply change there is a lot of issues, while Dahua seems to have plenty of certain chips (most likely non-HiSilicon) they are rejecting lots of orders on certain SKUs relating to HiSilicon.

Agree
Disagree
Informative: 1
Unhelpful
Funny

I have the 8 MP turret, model DS-2CD2386G2-I, and indeed it is miles better than the higher dollar "Smart" series from Hik in terms of AI performance. What you don't mention in this review, however, is that Hik's bounding boxes on identified people or vehicles are still very large at times.

Large enough to trip a line crossing line drawn on a driveway when the car is in the street, for example. If I recall correctly, your reviews on Bosch show bounding boxes much tighter to the object being detected, such that this would not happen. I would suggest that you need to include this sort of information in your testing. Hik's bounding boxes are also always rectangles, where again I recall seeing others with much more than 4 corners such that they can better follow outlines of objects. Unless I'm mistaken this fact alone means Bosch is better AI wise, and probably ditto for Avigilon.

The newer firmware is available at the link below, but I have not tried it yet to see if it makes a difference in bounding box size.

DOWNLOAD EU PORTAL

Agree
Disagree
Informative
Unhelpful
Funny
Subscribe to IPVM Research to read the full report.
Why do I need to subscribe?
The IPVM Research Service includes products tests and shootouts plus competitive and financial analysis, helping decision-makers better evaluate purchasing, partnering, developing, and/or competing against companies in physical security.
Already have an account?
Loading Related Reports