Hikvision Dual Lens Face Recognition Camera Tested

By Rob Kilpatrick, Published Nov 19, 2019, 09:35am EST

Hikvision's Dual Lens Facial Recognition camera, claims that it "adopts advanced deep learning algorithm and powerful GPU to realize instant face detection, capture, and comparison to accomplish face recognition and related alarm triggering."

We bought and tested the iDS-2CD8426G0F-I, examining:

  • Does it miss people walking through the scene?
  • Does it misidentify people?
  • Does confidence drop in different scenarios?
  • Does it recognize inanimate objects as faces?
  • What PPF is required for optimal performance?

****: **** ****** *** been **** ** ***** and ***** ***** ** the ***** *** * year ** **** *** was **** ******** ******** for **** ** *** United ******.

*******

*** ********* **** **** camera's ****** *********** *** fairly ******** ***** *********-*********** settings, **** ** ********* recognitions (*** ****** ********** as *******) ** **** a ***** ** *******, and ****** *********** ** WDR ****** ***** **** been ****** *** ***** face *** *******.

*******, ** ********** ******* inanimate ******* ** ***** (but **** *** *********), suffers **** ****** ***** and ******* ********** ** dark ******, *** ***** range ******** ** ***** systems ****** (~**-**') *** to *** **** ***** lenses **** ** *** of ********* **** ** fit ******** ******.

Key ********

** *** *******, *** Dual **** ****** *********** camera *** **** *********:

  • ** ***** ************:********** * ***** ** testing, *** ****** ***** misidentified *** ****** *** another ****** ********** *** scenes ****** *** ~*** faces ** *** ****** libraries.
  • **** ********** *** ***********:** * *****-****** ***** with *******, *** ********* dual **** ****** ********** subjects **** *** **** confidence ** ****** *** scenes. **** ***** *** historically **** ** ***** for ***** **** *********** cameras/systems, **** ** *****, AnyVision, *** *******.

******** ** *** **** some *********:

  • ********* ******* ******** ** Faces:******* ** *** ***** such ** ***** ** even ***** ******** ** floor **** ******** ** faces, ******* ****'* "**** Picture *********" **** **** useless **********. ***** ** no *** ** ****** specific ******* ** ****** them **** ********* ******* removing **** **** ** the ***** **** *** detection ****.
  • ***** ****** ** ****/** Scenes:** * ****/** ***** (~0.01 ***), ~*/* ** faces **** ****** **********, even **** * *** mounting ****** ** ~*.*** and * ******** ** ~10°.
  • **** ******** ******* *******/*********** Scene ** **/***** *******:***** *** **** ******** setting, **** ** ** IR ***** *** ***** may *** *** **** to ****** ***** ******* through, **** ****** ****** this ******* ****** *** camera ***********.
  • ***** **** ********** ******** Using *********** ********:***** *** *********** ** pixels (**** ***** ** pupil) ********* ******** *** only ~**-**' (~** - 104 ***). ***** ******* pupils *** ** ******* to * ******* ** 19, *** ********* ****** 10-11' ** **********.
  • ** ******** *********:******* * ******* ** a ******* ** *** camera ******** * ********, high ********** *****.

Vs ***** **** *** ******

*** **** **** **** recognition ****** ********* ****** than ******** **** ***************** ** *** ******* in *** *** ****:

  • ***** *** ***** ****** mistook *** ****** *** another ******* ***** **** two ***** ** *******, the **** **** **** Recognition ****** *** *** mistake ****** *** ******* person ** ~* ***** of *******.
  • ************, *** ********* ****** provides *********** ***** ~* lux *** ** *** integrated **, ***** ***** Face *********** ****** *** not *******.

*******, *** ***** **** camera *** *** ****** from ********* ****** ********* issues ** *** *****, a ******* ***** ** the ********* *****. *******, the ********* **** **** camera ** **** ********* (~$2,200 **. ~$*,***) *** requires * ******** ***** supply ******* ** ***.

Vs *********

*********'* **** **** **** rec ****** *** *** key ********** *************:

  • *****, ****** *** ****** is ********* ******** ** typical ** *******, ** is **** ***** ****** than **** * ***** AnyVision **************, ***** ****** at $*,*** **** *** channel, *** **** *** include ****** ******** ** cameras.
  • ******, *** **** **** camera ********* **** ****** in *** *** ** scenes **** *********. **** when ******* **** **** performing ****** **** **** and *****, *********'* ********** dropped *********** ** ***** scenes.

*******, ** *** *** drawbacks, ** ****:

  • ********* ******* **** ********* object ********* ******* ** AnyVision. *******, ** ******** no ***** ** ****** these ******* ** ****** them **** ****** ********** as ********* ****.
  • ******, *** ****** *** iVMS ** *** ***** any ** *** ******** search ******** ***** ** AnyVision, **** ** ********* images ** ** ******* in ******** *****, ****** for ******** ********, ** search ** ********* ****** cameras.

*******

*** ********* ***-**********-* *** be *********** *** ~$*,*** ***.

Physical ********

*** **** **** **** recognition ****** ***** ****** different **** *** ******* security ******, *** **** head ***** *** ******* which ** **** ********* to * **** *** containing *** ******* *** power ***********. ***** ** review ***** ******** ***************.

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

*** **** **** ****** did *** ******* *** person *** ******* ********** a ***** ** ******* and ***+ ****** ** the ********. ********** *** very **** ** **** scenes, **** ~**-**%, ********* low ***** (~*-* ***), dark (~*.** ***) **** IR, ***, *** ******* face ****** *** *********.

Screen Shot 2019-09-24 at 7 24 21 PM

Short ********* ***** *** ******* ***********

***** ** ****** ******* the ******, *** ****** recommended *******, *********** *** optimal ** ******** ****** ranging **** ~*° ** ~33°. ** ***** **** missed ***** ***** ******** in ****** *** ******.

*******, ********* ******** ** limited ** ~**-**' (~**-*** PPF) ****** *** ******'* 90° ***** ** ****. Subjects ******* **** ***** were ****** *** ********, making *** ****** ******** in ***** *****.

Screen Shot 2019-09-24 at 7 48 25 PM

Low ********** *** ****** ********* ** ****** *****

***** ***** ******* ***** distances ******* ** ****** range *********, *** *********** beyond ~**-**' *** ********, with **** ***** ********** scores (~**-** **. **-**). However, **** ***** ***** settings ** ***** *** no ***** ************.

Confidence *******/****** ** ****** ********* ** **

** * **** ***** (~0.01lx), ********** ***** ~** points **** *** **** is ******* ******* *** the ****** *** ******* downtilt (~*°). ******* ******** of ~**° ******* ** missed ********** ** *** confidence ***********.

Poor *** ***** *********** (~* ***)

** * *** ***** scene, ***** **** **** detected ** * *** confidence (~**-**) **** **** a *** ******** ******/******** angle *** ** ***** in *** ***** *********** the **** ** *** face. ***** **** ****** detected ** ******* ****** of *********, **** **** lower ******** ******.

Solid *********** ** *** *****

*** **** **** **** recognition ****** *** **** to ******** ********* ****** at "******" ********** (~**-**) in *** ******, **** as *** ******** ******* lobby **** ******* *****. Other ******* *** ****** recognition ******** ** **** tested **** ****** ** reliably ******* ***** ** this *****.

Lower ********** **** ****** *** ******* ***/**********

********** *** ********* ~** points ***** **** ******** were ******* ****/********** ******* through *** *****, *** recognition *** ***** ***** in **** *** ******.

********* *** ***** **** of *** ****, **** as * ***** ** mask, ******** ** * total ****, **** ** detection ** ***********, ********** of ********, *****, ** settings.

Inanimate ******* ******** ** *****

****** *******, *** **** lens ****** ******** ******** inanimate ******* ** *****. For *******, *** *** in *** ***** ** view ***** *** ******** constantly. ** **** *****, inanimate ******* **** ** a *********** ******* ** detections *** ***.

**** **** ***** ** no *** ** **** objects ** *** * face ** ****** **** in ****** ********** ** was ******** ***********.

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

*** **** **** ****** does *** ******* ******** detection. ******* ******* *** scene **** * ******* picture ** * *******'* face ******** * **** confidence ******** ******.

Face ******** ******* ******

*** ****** ******** **** exposure, ***** ** ******** to ****** ******** ** the ****** **** * face ** ********. *******, when ***** ** ** integrated ** ******, ******** were *********** **** ~**' distance, *********** ******** ******* and **** ********* ***********.

************, ******* ****** ** while **** ******** *** enabled ******** ** * totally ***** *****. **** image ******** *** ~* minutes ***** ******** ********** properly. [[***, **** *** hikvision *** ***** ****?]]

iVMS ***********

*** **** **** **** Recognition ****** ********** **** Hikvisions *** ****-****, ******** live ******* ** ***** captured *** ********** ***** with ****** *************.

Third ***** *** ***********

********* **** **** **** Recognition ****** ********** **** Milestone *** ************ **** *********** ******, ***** * **** recognition ***** ** ***** it ** **** **** in *** ******. *** plugin **** ****** ** image *** ***** ** the ****** ******* ******* the ***** **** *** alarm ** *********, ** does *** ******* *** person ******** **** *** library ** *** ******** metadata.

*** **** *********** ********* do *** ********* **** Avigilon, ***********, ******* ******** Center, ** ******* ***** NxWitness.

Some ** ******** ** *******

**** ** ******** ** the **** **** ****** were ** *******, ******* the **** **** ** bought *** ****** **** an ********** ** *********** and ******* ** *** latest ********* ***** ******** firmware.

*** *******, ***** *****, we **** ****** ** input ******* *****/******** *********** when ******** ***** ******* dropdowns *** ** *******:

Setup/Adding ***** ** *******

******* ** **** *********** is ******* ** ***** Hikvision *********. ***** ****** need ** **** *** detection *** ********* ***** (if *******) *** ***** the "******* ***** ********" (the ****** ******* ******). We ******* ******* ***** distance ****** *****.

Single ** **** ******

***** *** ** ***** to *** ******* ************ or ** **** ** selecting * ****** **** of ***** *****. *****, we *********** *** ******* of ****** ***** ** a ******* ** **** and ******** ** ***** recognition ** *** ******.

Country ** ******

*** **-**********-* *** ************ in *****.

Versions ****

*** ********* ******** **** used ****** *******:

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

Comments (15)

**** ** ********* **** access ******* ****?

*****, *** *** *** names & ***** *** be ***** ** *** PRC ****** ********

******* *** ** ******** to ****? *** ****** out ****, *** ** A ***** ****** ****** AND *** *** ******** TO ******* ****! ************.

*** *** *** ******** TO ******* ****!

#*, ****** *** **** feedback. *'** ********* **** at ****** **** -****** **** *** ******* Dahua *** ********* ******** Because ** ***** ****** Abuses?

***** ******* - ***** in-depth, ***********, ******** ***** is *** ********* *** is ******** ** ********** accurate *********.

****, * **** ***** for ***

*** *** ***** * hard **** *** **** admitted & ****** **** hikau ********.

*** *** ***** * hard **** ******* ******* the ***** *** ****** honest, ******* ******* ** these ********.

*** **** ** ** upsetting **** *** & anti ***** *****, **** means *** *** ****** certainly ***** ****** & objective **** **** *******.

** *** *** ** the *** **** ** a ********* ******** ******* with ******** ******* ** the ************ & *********** of ** **** *******.

**** ***** *****, ** Microsoft? **** **** **** accused ** ***** ****** violations. ****** ** **** using *** ******* *** Windows ***** *********?

** ******? **** ******** reports ** *** **** letting ***** ************ ********* workers *** *** ********, or ********** **** *** using *** ********?

* ** *** *******, or ********* **** ********* has ****, *** *** isolate ***, *** *** the ******? * *** this ** * **** dedicated ** ******** *****, but **** **** *** and ****** ******* ******** runs ** *******. **** of ** *** *** operate ***** ******** ** part ** **** **** and *** ** ********. Many ****** ******** *** all ***** ** ******** used *** ******** *** Amazon. *'** ****** ***** there.

***** ** *** **** the ****? *** **** HikVision? ******** ** *********, Google, *****, ** ******, they *** ****. ***** by * ********* **********, sure. *** *** **** done, *** ******** ** do, *** ******.

***** ** *** **** the ****? *** **** HikVision? ******** ** *********, Google, *****, ** ******, they *** ****. ***** by * ********* **********, sure. *** *** **** done, *** ******** ** do, *** ******.

******, **** *** ****** or ***** ***** ** any ********** (** ** PRC ** *********), ** commit ***** ****** ******?

** ******* ** **** your ****** *** ** used ** ******* *** organization ** ********.***** *** **** **** hundreds ** ******** ** dollars ** ***** *** operate ****** ******** ** Xinjiang **** *** **** to ******* *******. **** **** *** cross '*** ****' *** you?

* ***** * ****** clarify, ** "****" * mean ** ***** ** physical ****, ********, *****, etc.

********* ** ************ *******, ***** *** *** several ***** *** ****** Rights ********** ** **'* Global ****** *****, **** the ****** ** ********** growing. *** ** *** as **** ********* ** doing, **, *** ***** violations.

***** ***** *** ******* Microsoft** ***** ********* **** what ***** ** *****.

******* *********** ** ******* numerous ********* ** ***** Labour.*** * *** ****** admit ****'* * *** of * *******, ** Apple, *********, ******, *******, whatever, ** *** ** the ****** ** *** material **** ***. ****'* really ** *** ****** companies.

*** *** ** ***** examples ** *** ** what ********* ** *****? I'm *** ****** ** imply **** ******* ***** companies ** *** ****** that ** ****** ****** what ********* ** *****.

**'* **** ****** **** about **** ***** *** re-consider ***** *** *******. But *** **** *****? Is *** ***** * was ****** ** ****. So *** *****'* **** IPVM ** **** ********* products. *** **** *****? HikVision ***, ***. ***** also *** *** **** linked *******. * ***'* disagree **** **** *** bit. *** **** ***** the ***** ********* ********** lesser **********? *** **** single *** *** ** Dahua, *** *** ****** involved?

**** ****** *** ****** carries * ****** ****** sentence **** *** *******.

**, ** *****...

* ***** **** ********** it ** *** *******, but *** **** ****** were ** *** ******* during *** **** ******?

** *** * ******** of *** ***** ****** testing *** **** ********** faces **** ***** **** unknown ****** ******* ******* which ******* ** ** 108.

* **** ** ******* violations =)

** ***** ** *********** to **** ** **** how **** ***** *** be ********* **************.

*** *** ** ****. But * ** ******** it ***** **** ** processing *****, ***** ** why ********* ** ** costly.

Read this IPVM report for free.

This article is part of IPVM's 6,651 reports, 895 tests and is only available to members. To get a one-time preview of our work, enter your work email to access the full article.

Already a member? Login here | Join now
Loading Related Reports