Axxon Behavior Analytics And Facial Recognition Tested

By Rob Kilpatrick, Published May 26, 2021, 11:10am EDT

Axxon's Behavior Analytics claim to "Recognize potentially dangerous situations by detecting specific human postures" and their facial recognition is "free of recognition issues which were typical" in past generations. But how do they work in the real world?

IPVM Image

We tested Behavior Analytics and facial recognition for 2+ weeks, examining these issues:

  • How well do behavior analytics detect postures such as raised hands, sitting, or falling?
  • Do they accurately detect social distancing violations?
  • How do camera angles impact behavior detection?
  • How does it perform in real-world surveillance scenarios?
  • Does facial recognition 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 does it perform at night in low light (~2 lux) or dark/IR (~0.02 lux)?

*******

*****'* ******** ********* *********** was ***** ** *** testing, **** ******** **** as ****** ***** ********** detected ***** *** ****/**** detection ******** **** **** misses *** ***** ******, and ****** ********** *** sitting ********* ****** ********** only ** ******** ******.

*****'* ****** *********** ********* well ** *******, *** light, *** ****/** ****** when ***** **** ************, handling ***** ****** ****, with ** ***** ********* or ****** *********** ** registered ********, *** ********** and **** ************* ******* confidence (****** *** *** cause ******), ***** ***** reduced ********** *******, ******* frequent ******.

************, ***** ******** * separate *** *********** ** manage ********* (***** ****) and **** **** *** and*****'* **** ****** ************ *** **********, ** faces ** *** ********* cannot ****** ** ******** in ******** *****, *** can ******** ** ***** to *** ********* **** search.

****** *********** *** **** accurate **** ************* ** **** *********** **** **** *** not ******* ********. *******, age *********** *** ********** inaccurate, **** *******'* **** consistently ******** ** *-** years ****** **** ******, tested ** ******** ******* in *** **** ***** to **+ ***** ***.

*******

*****'* ******* ** "***-****", lower **** **** ********** facial *********** ********* **** as ********* ** ********, but ****** ************'* ********** ****** **** rec($***). ************, ******** **** as **** ********* ********** ************ ***** **** *** free ** ******* ***** platforms *** ******* (******* ********* ********)

***** ******** ********* *** Facial *********** *** ***-** modules *** *** ***** Next ************ ***/** ***** Next ******** ********. ***** licensing *** ** ******** (all ******** *** *********, not ************):

  • ***** **** ******** ****** License: $*** *** ****
  • ***** **** ****** *********** License, * ***** *******: $750 ***
  • ***** **** ****** *********** License, ** ***** ********: $5,000 ***
  • ***** **** ***** ******** Analytics (**** *********) *******, 1 ***** *******: $*** USD

*****'* ***** ********* ********* are ******** ** ** charge ** ***** **** Universe *******.

Behavior ********* ********

*****'* ******** *********, *** "skeleton" ** "***** ******" analytics, *** ******* ** detect * *********, ***** below. ** ****** *******, man ****, ***** **, active *******, *** "*****-******** people *********" (*** ****** distancing), *** **** ****** applications.

IPVM Image

*******, ***** **** **** Active ******* ********* *** People ******* (******* ******* masking) *** ** ****** marketed *** **** ** removed ** ****** ********.

Man **** ******** ******** ***********

** ***** ** ******** calibrate *** **** ******** analytics *** **** **** draw ******** ***** ********** the ***** ** **** the **** ** * person ** *** *****.

IPVM Image

Behavior ********* *** ********* ********

** *** *****, ***** reliably ******** *** ******* of *** ** **** arms (*.*., ***** ** in * *******), ***** here:

IPVM Image

Man ****/****** ******* ********* *** ********

*****'* *** ****/**** ********* was ********** ** *** tests, ***** ******** ****** when ******** **** ****.

IPVM Image

***** **** *** ******** alarmed ** **** **** because *** ******* *** far **** ********, ***** is *** ******* ** their **************:

** ** **** **** a ****** ********* **** the ********. ** *** experience, **** ***** *** rarely ***** ** *** implementations, ******** ** ** not **** **.

*** **** ********* **** failed ** ****** ******** falling ******** ****** ** away **** *** ******:

IPVM Image

***** **** ** **** case, *** ******** ***** not *** *** *******'* legs *** *** ****** to ********* ********, ************ a ****** ****** *****.

***** *** ** **** detected *** **** ********. And **** ** *** reason. *** ************** ** to *** *** ****** camera **** * ********* angle ** ****** *** whole ****.

*******, ** **** *****, falls **** *** ******** simply *** ** *** detector ********** *******.

IPVM Image

****** ******* ********* **** alerted **** ******* *** side ** * ****** sitting ****.

IPVM Image

** ***** *** ****** a ****** ******* ****** the ******.

IPVM Image

Social ******** ****** ***** ******** ** ******* ******

****** ********** ********* ************ alerted ** ****** ******* each *****. ***** ***** two ****** *** ~**' apart *** ***** ****** thinking **** *** **** than *' *****.

IPVM Image

**** **** *** **** by ****, ***** **** not *****.

IPVM Image

Active ******* ********* ***** ******

****** ******* ********* ***** alerted ********** **** * person ****** ***** ****.

IPVM Image

***** ******** ** **** this ** *** ******** anymore *** **** ** taken *** ** ****** versions.

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

***** **** *********** **** offers ****** **** ****** in *** ******* (********* or *********) *** ******** (shown *****). ****** **** other **** *** *********, it **** *** ***** users ** ****** *** a ******** ******* ** the ******* ** ****** a ****** ** ******* faces **** ** *****.

IPVM Image

***** ****** * ******** analytic ***-** (**** ******) ***** ****** ***** to ****** ***** ** recorded ***** ** ****** photos ** ****** (*** our ****). *******, **** Search **** * ********* UI *** **** *** use *** **** **** library ** **** **** rec.

***** **** *** ********* feedback **** ***** ** users ***** ****** ***** in *** ******* ** from **** *** ******:

**’* **** ******. ****** the *********** ******* ** was ******* ** *** Face *********** ** ***** Next ****** *** ****** purposes. ** **** **** you **** **** ** know **** * ****** belongs ** * ******** list. **’** ***** ** implement ********** **** ********** in **** ******** *** other ***-*****.

No ***** ********* ** ~* ***** ** *******

** ~* ***** ** testing, ***** **** ** false ********* **** ***** Axxon's ******* ********** (**) or **** ******** *** confidence ********* ** **-**. This ******** * **** library ** ~*** ****** and ~*** ****** ***** detected ** ******** *******.

IPVM Image

False ********* ********** ** **-**% **********

***** ********* ******** *** confidence ********* ** **-**% in *** **** ********* to ******* *********** ** low ***** ** **** partially ******** *****. ** our *****, ******** *** threshold ** **-**% *** not ***** ******** ******.

*******, ******** ** *******, to **-**%, ***** ************ trigger ***** ************.

IPVM Image

Reliable ******** ** **** ***** **** ** ***** ********* (~**°)

**** *********** ********** **** dropped ** ~* ********** points ** **** ***** (from **** ***% ** low/mid ***), **** **** cameras **** ******* **** with ***** ******** (~**°).

IPVM Image

**** **** ******* ** a ***** ***** ** the ****** ** **** downtilt, ********** *** **** reduced ** ~* ********** points.

IPVM Image

No ********* ** **°

***** **** *********** *** not ****** ***** **** people ******* ** *******, which **** ********* *** expected ********.

IPVM Image

Low ***** (~***) ********** ****

***** *****'* *********** ********** threshold, **-**%, ******** **** reliably ********** ** ~* lux *** ***** ***** 25% ********, ****** ********** was ******* ** ~** percentage ******.

IPVM Image

*******, ** ~* *** using ********* ****** (~**°), confidence ******** **** ******, dropping **** ** ~*-** points.

IPVM Image

Dark (~*.***) ********** ****

******* ** *** *****, recognitions ** *** **** (~0.2lx) **** ***** ******** with **° ******** ***** the *********** **-**% **********, although ********** ********* ******* by **-** ********** ******.

IPVM Image

** **** ** * shallow ***** ** ~**° confidence **** ******* ** ~10-15, ***** **** ***** the ********* ****** *** tests.

IPVM Image

***** ********:

*** ********* ***** *********, we ********* ** *** colored ******. *** ***** presence ******** *** *********** quality.

Sunglasses *** **** ************* ****** **********

* ****** ******* ********** was ***** ********** ** our ******* ***** *** recommended ********* ** **-**, even ** ********* ** ~33°, ****** ********** ******* by ~**-**.

IPVM Image

*********, ******** ******* **** were ********** ** ********* of **° **** ***** walking **° ****** *** camera, **** * **** in ********** ** **-**.

IPVM Image

************ **** ******** ***** a ****** *** ******* a *********** ** *** and **********, *** **** at ******* ****** ** 16° **** ********** ******* by ~**-**.

IPVM Image

***** **** **** ***** results **** ******** **** partially ******** *****.

Masks ***** ****** ***********

****** *********** ********** ******* by ~**-** ******, ********** completely ******* ****** ******** in *** *****.

IPVM Image

***** **** **** **** improved ****** *********** **** the **** **** *** will ******** ** ******* on ****** *****:

* **** ** *********** quality **** ******** * significant **** ** *** face ** * ****** situation. *** **** ***** these **********, ** *** trying ** ******* *** quality ** *********** ******* additional ******** ** ****** faces. **** **** ** possible ** ****** *** error ** *********** ** people ******* ***** ** ~ ** ***** **** the **** ****. ** the **** ****, *** error ** **** *********** without ***** ** *** times ****. ** ******** to **** ** ********* the ********* ** *** face ** * ********. We *********** ******* *** algorithm’s *********** *********** ******** conditions.

Separate ******** **** *** **** *******

** ***** *** **** recognition ** ******* *** user **** ******* *** set ** *** ***** Data ********, *** **** can **** *** ***** to *** ******** ** people ******* * *** interface, ***** ** *** below *****.

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

***** **** **** ********* consistently ****** ** ****** a ****** ******* * mask ** **** *** a *****.

IPVM Image

***** ********:

**** ** *** ****** here ** **** *** algorithm ******’* *********** ***** the ***** ****, *** the ******* **** ******. We **** **** ************ train *** ****** ** avoid **.

************, **** ********* ***** be ******* ** ******** the ***** **** ** a ******** **** **** his ***.

IPVM Image

***** ********:

***** **** *** ********* could ** ******* **** way ******* ** ******* only **** **** ***, in **** ****, ** looks **** * ****, please ***** **** *** out:

IPVM Image

~130-135 *** ****** *** ****** ***********

****** *********** ****** ~***-*** PPF ***** *** ************* higher **** ********* **** as ******** **** ****** or *********.

IPVM Image

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

***** ****** ********* **** not ******** ****** ****** when * ****** *** long **** ***** ** found **** ****** ********* to ** ** *** testing, ********* ** **** accurate ****** *********.

IPVM Image

***** ********* **** **** trained ** *** **** long ****:

*** **** *********** ****** is ***** ** ****** models *** *** ****-**** contain ******** ** *** with **** ****.

Age ********* **********

*** ********** *** **********, generally ******** *** *** higher **** *** ******'* actual *** ** *-** years.

IPVM Image

***** **** **** *** expected *********** *** **** age ********** *** **** for ***********/****** ********, *** specific ****:

*** ******* ** **** detector ** ** ****** statistical **** ****** **** individual *** ************ (***** may ** **********). *** instance, *** *** ********** feature ** ****** *** retail ***** *********** **** is **** ******* **** in **** ******.

Human ********* ********** ******* *******, ********, **********

** *** *****, ***** detection ********** ******* ** people *******, ********, ** obstructed.

IPVM Image

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

*******, ***** ********* ********** triggered ***** ****** ** trees ** *** ***** of ****, ***** ****:

IPVM Image

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

****** ** *** *** false ***** ** ****** indoor ******* **** ** major ***** *******.

IPVM Image

************, ** *** *** false ***** ** ******* in *** *****.

IPVM Image

*****

*** ** ** *** analytic ******** ******** *****, the **** **** ****** the ******** ** *** detection ******** *** **** setting ** * **** and ******* ** ** an *****, ***** ** the ***** *****.

******** ****:

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

  • *********: *.*

Comments (7)

*****; ****, **** ******* And ************* *******. ***** you.

"***** ******** ** **** this ** *** ******** anymore *** **** ** taken *** ** ****** versions."

*** **** ********* ** to ***?

Agree
Disagree
Informative
Unhelpful
Funny

***, ** ** ***** taken *** ** ****** versions *** ** *** prone ** *** ** false ******, * ****** just ******* ***** **** up ** ***** ** them ***** **** ****** trigger ** ****** ******* alert ** *** *****.

Agree
Disagree
Informative: 1
Unhelpful
Funny

"***, ** ** ***** taken *** ** ****** versions *** ** *** prone ** *** ** false ******, * ****** just ******* ***** **** up ** ***** ** them ***** **** ****** trigger ** ****** ******* alert ** *** *****."

**, * ******* **** since **** ********* **** a ******* **** **** would **** ********* ** identify * ******* ** the ******'* *****. ***** sense *** ** **** were **** ******* ** body ********.

Agree
Disagree
Informative
Unhelpful
Funny

**** ******** ******* *** reporting. * ******** ********* the ***** ******* - it *** *** ***** in *** ***** ** (or *'* **** *** missed **!) ******* *** prices **** *** ** annual ************ ** ********* licenses. *** *** ****** this?

Agree
Disagree
Informative
Unhelpful
Funny

**** *****, **** ** many ************ ****** ****** these ****. **** ******* is *********.

Agree
Disagree
Informative: 1
Unhelpful
Funny

****** *****. **** ********* me, * *** **** thinking **** **** ************ prices.

Agree
Disagree
Informative
Unhelpful
Funny

** ** *** ********* licenses. *** *** ** to ****.*** ** *** more ***********.

**********: **** ** * sales *** *** *********

Agree
Disagree
Informative
Unhelpful
Funny
Read this IPVM report for free.

This article is part of IPVM's 7,011 reports, 932 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