Gender Video Analytics Shootout

By Rob Kilpatrick, Published Dec 14, 2020, 08:09am EST

Using video analytics for gender identification is expanding but how well does it really work? And what personal features are gender analytics effectively using to make their identification?

IPVM Image

We tested gender analytics for over two weeks with five manufacturers' cameras:

Inside, we examine:

  • What physical features most impact gender classification?
  • What was the accuracy percentage?
  • How does hairstyle affect classification?
  • Does clothing style impact classification?
  • Does body type affect classification?
  • How do person and face detection impact accuracy?
  • How does PPF impact classification?

*******

***** ** *** *****, all ******* *** *** most ******* ****** **** length **** *********** ******, making ** ** *** the ******* ********* ** whether ******* **** ** classified ** **** ** female. *******, ***** ********* were ****** *************** *** routinely **** ******** **** an ******** ***** ***** not.

**********, **** **** *** the ***** ******, **** subjects ** ******* ***** and ******* ********** ** male ** ****** ***** on ***** *******, **** no ***** ***** ***** on **** ****.

***** ******* ****** ** manufacturer:

  • ********'* ** ******* *** Appearance ****** **** ******* weighted ****, *********** ******** with **** **** ** female *** ***** **** as **** ** **** rates. ***** ******* **** as **** ****, ****** features, *** ******** ***** were *********** *******.
  • ***** *** ****** ********* similarly, **** ******* ********* hair, *** **** ******** style, *********** ******** ** skirts ** ****** **** often. ************, **** * face *** ********, ****** features **** **** ******* considered.
  • *********'* ****-**** **** *** camera **** ********* ****** analysis **** * **** was ********, ****** ****** which ********* ******** ** person *********, **** **** subjects **** ****** ****. Because ** *** **** PPF ******** *** *** camera ** ******* * face (~*** *** ** higher), ** *** **** likely ** ********* ******** gender.
  • ******* *********** **** ********** hair ** *** ***** and ******* ******** *****, facial ********, **** ****, etc. ******* *** **** the **** ************ ***** sometimes *** *** ******** gender, ****** ********* ******** as * ****** ******* of **** ** ******.

Accuracy ********

******** ** ****** ** be ***** **** (**%+) in **** ************, ** men ********* **** ******* hair **** *****, ***, expect ******** *** ****** who ** *** *** this ********** ***, ** applications ***** *** ********** does *** ** **, accuracy ** ****** ** be ***** ****.

Performance ********

***** ** ** ******** of *** *********** ** all ** *** ****** recognition *********.

IPVM Image

Avigilon *** ******* ********* **** ******

**** ****** *** **** heavily ******** *** *************** in *** ******* ** all ************* ** ********* (discussed *****).

******* *** ******** ******** hair *** ********. ** a ****** *** ******** with **** **** **** were **** ****** ********** as * *****, ********** of ***** *******, **** as ******* ******* ****** hair ** **** *****, shown *****.

IPVM Image

*********, ******** ********** ****** effectively ******* ***** ******* and ********** ******** ******* weighted ** **** ******.

IPVM Image

Dahua / ****** ****** **** ******* ** ****** *********

****** ****** ******, ***** and ****** ******** ****** on **** ****** ********* and **** *********, **** face ********* **** ********. When ***** ****** *********, hair ** **** ******** weighted.

*** *******, ***** ******* weighted ****** ** **** length **** ***** ****** detection, ***** ***** ********** as ****** ** **** was **** *** **** versa.

IPVM Image

****** **** ******* ******** people ** **** ****** when ***** ****** *********, resulting ** ******* ***** classified ** ***** ** hair *** *****.

IPVM Image

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

**** ***** **** ********, Dahua *** ****** ****** classification ****** **** **** accurate.

*** *******, *****, ****** considers *** **** **** long **** ** ** a **** **** * face ** ******** *** a ****** **** **** a ****** ** ********.

IPVM Image

*********, ***** ********** **** person ** * **** when * **** ** detected *** * ****** when *** **** ** detected.

IPVM Image

Body **** ******** **** **** **** ** *** *************

**** *** ****** **** types **** ******** **** and *********** **** **** short ** **** **** on * ******. ******, Dahua, *******, *** ******** would ***** ******** ******** as **** ** **** was ***** ** **** back, **** **** ***** female *******.

IPVM Image

Gender *********** **** ******** ** ***** ***

******* ****** *** ******** both ****** ****** ** long ****** / *** PPF ******, ****** ************** was ********** ** ****** distances ***** ******* ******** cannot ** *********. *** example, *** ***** ***** below *** ********** ********** as ****** ****** ** the ******, *** **** at * ********, ******* visible **** ****.

IPVM Image

***** ********* ***** ** get * ****** *** when ********* ****** *** still ***** *** ** guess ****** ** ***** PPFs, ********* ** ***** classifications.

IPVM Image

Skirts ******* ******** ** ****** *** *****

****** *******, ** * person *** ******** ******* a *****, ***** ***** automatically ******** **** ** a ******, ***** *****. Other ************* *********** *********** clothing *****.

IPVM Image

************, ****** ******** ****** more **** ****, ** a ********* ***** * female **** ***** **** was ******** **** ****** it ***** ********* ** classified ** ****, **** a ***** *** ****, the ******* *** ****** classified ** ******.

IPVM Image

Verkada **** ** **************

******* *** *** ******** gender ** *** ****** that **** ********. ** some *****, * ****** was ******** *** ***** to *******'* **** ** people ******** *** *** given * ******. *** other ********* ********** ****** whenever * ****** ** face *** ********.

IPVM Image

Hikvision **** *** / ******* ***** ********

********* ****** *********** *** the **** ******** ** our ******* *** *** also *** **** ***********, only ******** ***** ** be ******** **** ** and ** **** *** (~90+).

IPVM Image

Hikvision **** ***** ******

********* *** *** ****** hair ******* ******** ** other *********. **** ** long **** *** ******* with *** *******'* ****, they **** ********** ********** as **** ***** ****** would ******** **** ** female.

IPVM Image

Versions ****

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

  • ******** *.*-***-***-**: *.**.*.**(*****)
  • ******** ******* ******: *.**.*.**
  • ***** ***-*********-***: **.***.*******.*.*, ***** Date: ****-**-**
  • ****** ***-******: *.**.**
  • ********* ***-**********-*: **.*.** ***** 180316
  • ******* *****: *** *****

Comments (30)

David Cochard

12/14/20 02:32pm

**, **** ** *** think ***** *** **** that ******* ** *** classifying ****** *********? *** underlying ******** **: ** video ********* ***** ****** their ****** ** ******* a ***** **** **** the **** ** *********** (with **** ***** ** error), ** ******* ** not ******** *** ********* that ** *********?

******.

John Honovich

IPVM | In reply to David Cochard | 12/14/20 02:37pm

*****, ****** *** **** first ******* *** * good ********. ** ***** on **** ********, **** are ****** ********* ** use ****** ********* ***? Is ** ********* ************? Or ** ** "**** me *** ******* *" or?

* *** ** **** seems ** ** **** entertainment (*.*., *********** ** demo) **** ****** ***. To ** *****, * am *** ******** **** some ***** **** *** of ****** *********** *** I ******* **** ** not.

Avatar

Ethan Ace

IPVM | In reply to David Cochard | 12/14/20 02:57pm

* ***** **** *** classifying ****** *** *** time ***** ** ****, but ** ***** ** better ** ***** *** a ******** "*******" ********, instead ** ****** *** classifying. ** ** ****** now, *** *** ****** for **** ** ******, but ** *** **** to *** ***-********** ****** detections *** *** **** see **** ** ******* all ******* ***, ** you're ******* ** ******, male, *** ************ *** at ****.

David Cochard

In reply to Ethan Ace | 12/14/20 03:54pm

****** *****. * *****, the ****** ****** ******* data **** **** ********** and **** **'* *********, just *** **. **** like * ******* ** an ***** ***** ** sometimes **** ** ******* information **** * **** level ** ******** *** sometimes ***** ***. **** bad **** ** ***** than ** ****. ******** the ***** ** ****** specifically *** "*******" **** becomes ********* **** **** approach.

Avatar

Brian Karas

In reply to Ethan Ace | 12/14/20 04:41pm

** **** **** ***** an "************" ******** *** some **** *******, **** as ******, *******, **** covering.

** *** **** ********* used ** ****** ******* of ******, *** **** being **** *** ***** with * ****** ** specific ** ******** (****, with *******, **** *****, moving ** * *********), and **** **** *** some ** *** ********** depending ** **** *** are **** ** *** from * ***** *****.

*** ***** ****, ** course, ** **** ** have ****** ****** ** make * ********** ******** on ***** *******, *** the ******* **** ***** points *** ** **** many ***** *** ****** make * ******* ********, so ** *** ******* "Undetermined" ** * *** to ********** **** *** product ***** ** ** unable ** **** * hard ********. **** *** be *** ** *** resolution, ******* *** **** lighting, **********, ** ***** factors.

IPVM Image

IPVM Image

**** **** ** **** a ****** ******** *** both **** *** ******, where ****** ***** **** general ********** ******* **** consideration (********* ****).

** ** *******, *** can *** ** **** scene *** ******* ********** my **** ** ************ gender, *** ** **** to *** *** ********** and *****, *** **** using * ****** ******, accurately ********** ** ** Male (****** ***-****** ***** here).

IPVM Image

*******, * ***** **** gender ************** ** *** of *** ***** ***** we (*** ******** ** well ** ******) ***** likely *** * *** of *********** ** *** next ******* *****.

Avatar

Hash Salehi

Motorola Solutions
In reply to Brian Karas | 12/14/20 05:22pm

********** ********** ** ** Male

**** **** ********** **********? :)

Avatar

Brian Karas

In reply to Hash Salehi | 12/14/20 05:28pm

** *** **%, *** then * *** **** testing ***** ******* * Pumpkin ***** ***** *** it **** **** ** 51%.

Avatar

Hash Salehi

Motorola Solutions
In reply to Brian Karas | 12/14/20 05:29pm

***** *** ** ****** purchases

Undisclosed Manufacturer #8

In reply to Brian Karas | 12/14/20 06:44pm

********* ********** *******!

Undisclosed Integrator #10

In reply to Brian Karas | 12/14/20 09:12pm

************************** ***. ****. *** I **** *** * few ******* ***** ****** myself ** * ****.

Undisclosed Integrator #1

In reply to David Cochard | 12/14/20 03:13pm

* ***** **** **** is ********* ** ******* for ** *** ****. I ***** ****** ************* have * ******** *******/******* than ** **** ** question **’* ********. * know ***** ****** ** any **** ** ****** contributes ** * ******* that ***** ** ****** any ****** ** **** system(s). * ***** ******* and ****** ** **** ilk *** **** ****** providing **** ********** ******** versus **** ******* ** their *********. ****** ********* seem **** * ***** category. * ***’* ****** what ***** ** *** get **** ** ******** classifying ******.

Michael Miller

In reply to Undisclosed Integrator #1 | 12/14/20 03:53pm

*** * ** ******* for * ****** **** a *** ***** *** blue *****. ***** **** to ****** *** ******* for ***** ************ *** be ****** *** *** sure *** **** ****** are ***** **** *******. This ** ********* **** can ** **** **** Avigilon's ********** ******.

David Cochard

In reply to Undisclosed Integrator #1 | 12/14/20 03:59pm

* ***** ** *** all ***** **** * system **** ***% ******** would ** *** ****. But ***** **** **** will ******** ***** ****** (for * *** ** reasons), *** ******** ** to **** **** ** the **** **** ******. It *****'* **** *** system ******* ** ** solves **** ******* **% of *** ****. ******** search ** * ****** usage ** ****** *********.

John Honovich

IPVM | In reply to David Cochard | 12/14/20 04:16pm

** *****'* **** *** system ******* ** ** solves **** ******* **% of *** ****.

*****, ** ***** '****' 90% ** *** ****, as **** ** **** everyone *** ******-************* ********.

IPVM Image

****: ** ***, * think **** ***** ***** people *** ***** ****************. It's *** ***** *** computers ** **** ********, it's ******* ** *** viewer ***** ** ** and ****** **'* ********* wrong *** ******** *********.

Olivier HALLOUI

In reply to Undisclosed Integrator #1 | 02/23/21 03:02pm

" * ***’* ****** what ***** ** *** get **** ** ******** classifying ******."

** **** ****** ** a ****** ******** ****** field ** **** ***** be ****** *** ********* strategy *******, ** **** countries ** ****** *** mixity ******** *** ******** reasons, ** ** ******* if * *** ******* to ******* ** * women **** ****...

Undisclosed Manufacturer #2

12/14/20 03:20pm

** **** ******* ****** with **** * *** years ***. ** ******** misidentification *** **** *** system ********** *** ***** Leader (* **** ** his *** **'*) ** a ******. ** *** some ******* *** ** appears **** *** ********* gave **** ***** ****** to *** ****** *****. He's **** **** **** our ******* *** ***** 2 *****, *** * still **** ******* ***** it.

Avatar

Sean Nelson

Nelly's Security
12/14/20 03:43pm

*** *** **** ** post ***** ** **** topic *** *** ******** involved, ** *'* *** even ***** ** ** it.

Undisclosed End User #6

In reply to Sean Nelson | 12/14/20 06:40pm

*** ** *****. ** one *** **** ** tell ** **** ******** to ***.

Undisclosed #3

12/14/20 04:53pm

* ***'* ********** *** there ** * **** to ******** ** ***. I **** **** ******* manufacturers *** * **** while **** *** *************** cheapen *** ******* ********* package **** **** ** not ****. ***** ** works **** ********** * feel **** ****** *************** are **** * "****, why *** **** ***" feature **** *** *** for, *** ****, *** does **** **** **** good.

*** *******, * *** years **** ** *** a ***** ** ** otherwise ********** ****** *********** platform **** *** ******** rejected. *** ****** *** not ******* *** ****** recognition ******** *** ******. The **** ** ******* identification *** **** **** for ****. *** ******* reason ** **** * single **** ******* **** member *** ***** ********** as *'* ***** **** this ****** *** * 5'11" ****. **** ****** was *** *** **** cause. **** ***************** ****** a ******* **** ** the ***** ** *** the ****** ***** ***** of *** *******. ***** this *** *** *** sole ****** ***** ************* it *** ********** *** wrong ******. *** *** and ****** ************** **** has ***** **** ******* from *** ****** *********** analytics *******.

Avatar

Brian Karas

In reply to Undisclosed #3 | 12/14/20 05:05pm

* ***'* ********** *** there ** * **** to ******** ** ***

*** *** **** *** cases * **** **** are *** ****-***** ****** and *** ****** ********* applications (***** *** **** primarily ****-*****).

***, ** *** ** the ***** **** ** the **** ***** ** analytics, ***** ******* **** be **** ** ******* more ******* ***** ********** attributes ** ** ******, and **** ***** ** figure *** **** ******* of ** ******* *****, not **** **** ***** are ******** ******* ** the *****.

* ***** *** **** a ****** ***** ****** based ** * ****** for ****** ***************, ** I ***** ** ** unlikely ** **** ******** enough **** ** **** to ******* * ******** focused ** **** *** specific *** ****. *** when *********** *********, ** can ******* ***** ************* for *** ****.

Undisclosed Manufacturer #4

12/14/20 05:49pm

****** *** *********** ** the ******* ** *** wacko ****** *** ****** after ***. *** *** test * *****?

Avatar

Hash Salehi

Motorola Solutions
In reply to Undisclosed Manufacturer #4 | 12/14/20 05:56pm

***** **** ******* ** rather ***********, * ** think ***** ******** ******** to "**********" **** ****** identification ***** ** * better **** **********.

"**** **** ******* * red ***** *** **** jeans" ** ********* **** can ** ******** **** a ******* ******** *** less **** **** "* woman ***** ** ******."

Undisclosed #5

In reply to Hash Salehi | 12/14/20 05:59pm

* ***** ***** ** wallet

* ******** ****** ********?

Undisclosed Manufacturer #4

In reply to Hash Salehi | 12/14/20 07:37pm

*****, ** *** ***********. Should **** ******* ***** before ******* ****.

Undisclosed Integrator #7

12/14/20 06:42pm

😂*** *****. ***** ** one **** *** **, i ****. *** **** comes ** **** ** that ******* ***** ** the ***/*** ** ******** throwing * *** *********** as “**’**” 😂😂😂😂

***** ** ** ******** to *** **** **’** thru *** ********* ** see **** ** **********? 😂

Undisclosed Distributor #9

12/14/20 08:14pm

*** ****, ** *** system **** ** ***** (in ***** ****) **** will ***

Undisclosed Manufacturer #11

12/14/20 10:08pm

*** ******* *** *** mentioned *** *** ****** for **** ***** ********* based ** *** ******* in ***** ** *** monitor.

** *********** ***, ******, race *** ***** ******** “style” **********, *** ***** display ******* *** ** thousands ** ********** ********* messages.

** ** **** ** would ********* ********** ***********, golf ***** *** ******* pads.

** ****** ** *** person ** *** **** available? ****** *** * friend.

Hamish Dobson

12/16/20 02:52am

*** ********* ******* ***/****** as * ***** ** the *********** ****** *******. With * ****** *********** function **** *********** ******, the ******* ***/****** ****** only **** ** ** good ****** ** ******** a ****** *** *** a ********** *** ** images **** ***** ** further ****** *** ******. Once ** ***** ** selected **** ******* ****** results *** ****** ******* converges. *** ****** ** success ** ******** ** not ** **** ******* accuracy *** ** *** quickly ** ******** *** find **** **** *** looking ***. *** *** same ******, ****** ******* are **** **** ******** in ** ********** ****** that ** ********* **** live ** ******** *****. As *** ********** ******** such ** ******* ******** search ** *********** ********, more ******** ****** ****** will ** ****.

Undisclosed Manufacturer #11

In reply to Hamish Dobson | 12/16/20 04:51am

** *** ******** * search *** ** ****** key *********, ******** *** search ********* ***’* **** it **** **** ** has ******* *******.

* ****** *** ******, red *****, ***** *****. 20% ** *** ***** in *** ***** *** missed. **** ******* ******** it ****?

Humberto Macedo

12/22/20 03:02pm

*****. ****** *********** ********** either **** ******* *** mathematics ************ ** *** NOT ****** *********** **********. That ******!

Read this IPVM report for free.

This article is part of IPVM's 6,805 reports, 913 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