ISS SecurOS Facial Recognition Tested

By Rob Kilpatrick, Published Jul 21, 2021, 10:09am EDT

ISS claims their SecurOS FaceX facial recognition provides "extraordinary recognition accuracy", but how does it work in the real world?

IPVM Image

We tested ISS SecurOS Face examining:

  • How does it perform in real-world surveillance scenarios?
  • Does it falsely identify a person as a different person?
  • Does it miss faces walking through the scene?
  • Can it still recognize the faces of people wearing sunglasses, hats, and masks?
  • How accurate are demographic analytics like age, gender, hair color, and others?
  • How does it perform at night in low light (~2 lux) or dark/IR (~0.02 lux)?
  • Does it detect anything other than human faces?
  • Can liveness detection be spoofed with paper printouts and mobile phones?

This is our third test report on ISS' SecurOS analytics. Readers should also see: ISS SecurOS Person, Vehicle, And Animal Tracking Analytics Tested and ISS SecurOS Auto LPR And Car Make/Model Recognition Tested

*******

** ~* ***** ** testing, ******* ***** *** not *********** *** ******** using * ********* ** ~120 ****** **** **** 100 ****** ****** ******* through *** *****, ********* harsh ****** (~**° ********/~**° horizontal). *******, ** *** light *** **, ******** were **** ********** **** walking ******** ** *** camera, **** ********* **** tilts (~**-**°).

******** ******* **************** **** ********** **** confidence ******* ** ***** not ******* ****. *******, subjects ******* ***************** ** ******* ***** were ********** **** ~** percentage ****** ***** ********** or ****** *******.

*********** ********* **** **********, frequently ********* ***** *** men, *** ***** ***** as ******, ********** ** hair ******, ******** *****, or ***** *******, ****** adult *** **** ************ correctly **********. *** *** estimated ******* ~* *** 30 ***** ****** **** actual, *** **** ***** and ***** ********* **** often *********. *** **** physical *************** ************ ******** accurately **** **** *** glasses.

*******, *****'* ******** ********* accurately ******** ******** ** spoof *********** ***** ******* photos ** ****** ********* on ****** ******.

High ******* ****

*** ***** ** ********* (MSRP $*,*** ***) ******** to **** *********** **** rec *********, **** ** Avigilon ($*** ****) ** Briefcam (~$***-*** *** *******, 100 ******* *******), ****** this ******* ** ******* to ******, **** ** Herta ** **** (**** ~$2,500).

Compared ** ******** *** ********

** *** *****,********'* ********** ************************ ******* ***** *********** better **** ******* *****, recognizing ******** **** *** side (~**° ***** ** incidence), ******** ** ~**° in ***.

************, **** **** **** to ********** ********* ******** wearing ***** ** ***************** **** ****** ********** drop.

*******, ********** ****** **** rec ******* ** ***** than ***, ** $*** USD **** *** ******* compared ** $*,*** ***, while ******** ***** ** software **** *** ~$***-*** USD *** *******, **** a *** ******* ******* order (~$**,*** **** ****) or * $**,*** *** "starter ***."

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

** **** * ***** of *******, ** ******** were ************* ** ****** or ******* *******, ***** a ******* ** ~*** people **** **** *** people ******* ******* *** scene, ********* **** *********, staff ** *********** **********, and ********.

IPVM Image

*******, ***** ********** *** relatively ***** ******** ** most ************ ****** *********** is ****** ** ** implemented, **** ** ******* with ******* ******* ********, plus ***** *** *******, casinos **** ********* ** visitors, *** **** ******* with ********* ** ****** subjects. ******* ** ****** populations *** ****, **** false *********** ****** **** likely.

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

** * **** ***** setting ******** **** ********** approaching ** **° (**********) even ** ***** ********* of **° **** ** confidence ****.

IPVM Image

Subjects *** ********** ** **°

*******, ******** **** *** recognized **** ******* **° across *** *****, ******** in **** *********** ******* such ** ******** ** Briefcam.

IPVM Image

Significant ********** **** ** *** *****

******** **** ********** ** a *** ***** ***** scenes **** **** *********** the ****** ********-**. *********** was ******** ** ** a ******** ** ~**-**°, though ********** ******* **-** percentage ******.

IPVM Image

Drastic ********** **** ** ****/**

*********, ** * **** setting **** ** ** (~0.02lx), ******** **** **** identified **** *********** ******** on ** ** **° of ********, **** ********** dropping ** ~**-** ******.

IPVM Image

*** ******** ** *********** in * */* ***** is ** ******* * good ********* *****, ******* the ********** ***** ** enrollment. ***** * ****** reference ***** ******** *********** somewhat, *** ***** ******** in ~** ********** ****.

Slight ********** **** ******* *********

******** ******* ******* * backlit ***** **** ********** accurately **** **** ~** confidence **** ******** ** an ****** *** *****.

IPVM Image

People ********** ********** ** ********** ** ****

****** ******* ******* *** scene ******* ********** **** still ********** ** ** ~33° ** ******** **** only ~*-** ********** ****.

IPVM Image

*********, ******** ******* **** were ********** **** ** harsher ********* ** **° with **** ~*-** ********** lost.

IPVM Image

Subjects ******* **** **** *** ******* **-** ********** ****

******** ******* **** **** and ******* **** **** identified ** ******* ****** of **° **** ****** a ********** **** ** ~40 ** *** ********** at ***.

IPVM Image

Drastic ********** **** *****

******** ******* ***** **** frequently *** ********** ** all.

IPVM Image

**** **** **** ********** while ******* ***** ** was **** ~** - 60 ********** ****.

IPVM Image

*** ********* **** **** expect * **** ** confidence **** ***** *** worn *** **** **** recommend ******** ** ********** to **** *** **** under ***** ***** *** for ******** ***********.

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

********* ******* **** ** car ****** *** *** populate *** **** ********** tab, ***** *** **** a ******** ** ***** face *** *********.

IPVM Image

Liveness ********* ******** ***** ***********

*********, ***** *** ******* of ** *** ******** detection, ***** ******** **** not ****** ** *** scene ***** * ******* of * ****** *** to *** ******** ***** being **** *** *********.

IPVM Image

***** **** **** ******** attempts *** ********* *** liveness ****** * ** spoof ********* ****** *** spoofing ********.

IPVM Image

******* * ******* ** a ****** ** ** a ***** *** ****** with ******* ******** *** to * ********** * liveness *****.

IPVM Image

*** ******** ** ******** is ** *** *** threshold ********* ** *** scene, **** **** ******* scenes ********* ****** **********. They **** ********* **** this ******* **** ** improved ** ***** **** release.

Mask ********* ******** *** ******* ** ******** ***** ****

**** ********* ********** ******** when ****** **** *** wearing ***** **** ******* through *** *****.

IPVM Image

****** **** ***** ** spoofed ** ******** *** lower **** ** * person **** **** ***** arm.

IPVM Image

************, **** ******* ******* drinking * *** ** coffee ** ******** *** lower **** ** * face **** * **** spoofed **** *********.

IPVM Image

Gender, ***, *****, **** ***** ************ **********

************ **** ** ****** recognition **** **********, **** most ****** ********** ** male, ********** ** ****, clothing, ** ***** "*******" features. ************, ******* **** were ************ ********** ** female.

IPVM Image

*** ********** *** ************ higher **** *** ******'* actual ***. *** ******* below ** ************* ** ~10+ *****.

IPVM Image

*** ********** *** **** less ******** ** ******** with ******. *** *******, the ******* ***** ** overestimated ** ~** *****.

IPVM Image

***** ********* ******* ****** without ****** ** ****** with ****** ********** *** even ******** ****** ****** as ******.

IPVM Image

**** ***** ************ ********** got **** ***** ***** and **** ********** *** color ** * **** person.

IPVM Image

Hat *** ******* ************ ********

****** ******* **** ** glasses **** ******** ********** as ******* ****.

IPVM Image

*** ******** ********* ************ is **** ** ** currently ***** ********** *, it ** *** ********* their **** ***** *** they *****'* *** **** use ***** ** ****** using ********.

*****

***** ******** ********** *** correct ******* *** ********* a ******* ** ******, shown ** *** ***** video.

Versions ****

*** ********* ******** **** used ****** *******.

  • *******: **.*.***

Comments (4)

*** *** **** **** a ******** ******** ** the ***** ******** **********?

Agree
Disagree
Informative
Unhelpful
Funny

***, *** *** ********* on ****? **** ** you **** ** * relative ******** ** *** population?

Agree
Disagree
Informative
Unhelpful
Funny

** *** **** ********* home ***** *.*. ** it ********* *** ********* in-house ** *** **** still ********* ********'* ********* or ** ** ****** domain **** ********? ** something ****?

Agree
Disagree
Informative
Unhelpful
Funny

* ********* **** ******** to *** *** **** responded **** *** *********:

** **** *** ****** with ******** ** **** years. ********* ******* ** developed **-*****.

Agree
Disagree
Informative
Unhelpful
Funny
Login to read this IPVM report.
Why do I need to log in?
IPVM conducts unique testing and research funded by member's payments enabling us to offer the most independent, accurate and in-depth information.
Loading Related Reports