Avigilon Appearance Search Tested

By Rob Kilpatrick, Published Oct 30, 2019, 10:32am EDT

Avigilon Appearance Search claims that it "sorts through hours of video with ease, to quickly locate a specific person or vehicle of interest across an entire site, or multiple sites." But does it really save time compared to typical search?

free image2 2

To find out, we tested these claims and others, examining:

  • Is it faster than searching through footage on a normal VMS using timeline or thumbnail search?
  • Is searching by specific colors accurate?
  • How accurate it is in full light, low light, and IR?
  • Does it accurately classify age and gender?
  • Are vehicles accurately classified and searched?

Executive *******

***** ** *** *****, Appearance ****** ******** * significantly ****** ***** ** search *** * ******** person ** ******* ****** multiple ******* ******** ** using ******** ** ********* search *****. *******, ***********/******** identifiers **** ** ***, gender, *** **** ***** were ******* ********** ** our ***** *** ********** IR (*** ********* ******) ****** ************* ** objects *** ****** *****/****.

Key ********

** *** *****, ********** Search *** ******* *********:

  • ****** **** ******* *** search:***** ********** ****** *** significantly ****** **** ********* through ********* ** ***** typical ********* ******, ********** across ******** ******* *** multiple ****.
  • ******** ***** ****** *******:********* *** ***** ** lower **** ***** *** generally ******** ** *** tests ** ******* *** moderate *** ***** ******.
  • ******** ******* **************:********** ****** ********** ********** and ******** ******** ******* types, ********* ****, ******, buses, *********** *** ********.
  • ****** *************:** *** ********** ******, users ****** ***** ** box ** ****** ** on **** ******. ** further ************* ** ********.
  • ** ********** **** **** ACC:********** ****** ** ******** with ******** ******* ****** Enterprise ** ** ********** cost. ** ******* ********* is ********.

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

  • ********** *********** ***********:********* ** ****** ** age (*****/*****) **** **********, with **** ******** ***** as ****** *** **** versa *** ****** ***** as ******** *** **** versa.
  • **** ***** ** **** hair ******:**** ********* ***** ** black ****, ******** **** hats **** ********** ***** in ****** *******.
  • ********** ** *** *******:** *** **** **** IR **, *** ******* effectively ****** ***** ** gray, ******** *** ****** they *** ** ********. Results ** ******** ******** in ** **** **** less ******** **** ***** in ***** ******.
  • ********** **** *******:** *** *****, ********** Search ********* ******** ********* objects (***** ** *****) or **** **** ** a *******'* **** **** adding ** *** "**** Profile." ***** ******* ****** be ******** ******** ** disregarded.
  • ******** ********* ****:********** ****** ***** **** with ********'* *** *********, not ***** *******, ************/********** ** *** ** their ********* ********** (* significant ********** ****).

** *********** **** ** these **** *** **** of ********** ****** ** this *****:

Accurate ******* ******** ***** ******

** *** *****, ********* upper *** ***** **** clothing ****** *** ******** during *** *** *** in ******** *** ***** (~1-5 ***). ******** ****** were **** ******** (***, yellow, ******, ****) *** brown, ****, ***** **** accurate, ** ****.

Search For Upper Lower Body Color Accurate

Subject ****** ********* ********

********* * ****** ** a ******** ******* **** recorded ***** *** ********* accurate ** *** *****, with **** ******* ******* the **** *******. *** example, ******* ***** **** the ******* ** *** different ******* ** ***** different *****.

Specific Person Search From Recorded Video

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

********** ****** ******** **** detection ** *** ********* of ****** ******. ** a ****** *** ******** a **** ********** ******, it ** ***** ** the *******'* ********** ***********, shown *****. ** *** tests, **** * **** was ********, ******** *** generally ******, ********** ****** multiple **** *** ******** changes. *** *******, *** subject ** ******* ***** different ******/**** ** **** image:

*******, **** *** ** confusing *** ***** ** there ** ** *** to ******** ****** ***** faces ***/*** *** *****, there ** ** ********** of ******* * **** is ** ** *** detected, *** ***** ****** be ***** *** *****. This **** ********** ** ** ********** parameter ** *** *******'* appearance, ******* ** ******** color, ******/***** *****, ***.

************, ***** ** ** way ** **** * face ** "*****" *** tell ********** ****** ** exclude **. **** *** be ** ***** ** the ******** ************ ******** inanimate ******* **** ** the ***** (**** *****) or *****, ***** ****** be ******* **** *** subject *******.

Floor Detected As Face No Way To Mark Incorrect

Searching ***/****** **********

**** ********* ***** ********, searching ** *** *** frequently **********, ******* ****** adults **** ********* *** children.

Searching For Children Displays All Adult Subjects

************, ********* ** ****** was **** **********, ********* mostly **** ******* **** high ********* **** ***** were ********.

Searching For Female Displays Several Men

Searching **** ***** **********

********* **** ***** ******** false *******, *** ******* searching ******* **** ***** hair ***** ******* ******* of ****** ******* ***** hats, ********* *** ****** people ***** **** ******.

Search For Dark Hair Displays Subjects In Hats

Accuracy ******* ** ** ******

********* *** ******** ** dark ****** **** ********** IR ** ******** **** accurate ******* ** *** subject's *****(*) *** *** be **********. *** *******, despite ***** ****** ** the ******* ** ** below, **** ********** ******* of ****** *** *****.

Accurate ******* ****** *** ***** / ****

** ******** ** ******* clothing, ******* ***** ****** also ****** ****, **** red **** ***** ***** over *** ****** ** several ****.

Vehicle Color Accurate

******* **** *** **** accurately *********, **** ****** and ***** ************** **** cars:

Trucks Accurately Classified Searched

*** *********** ********* ********** differentiated **** ******** ** other ********:

Motorcycles Accurately Classified Searched

Simple *****

** *** ** ********** Search, ********'* ********* ******* must ** ********** *** installed ** **** ****** (~5 ****** *******). ***** installation, ********** ****** *** be ******* ** ******* cameras.

Pricing / *************

******** ********** ****** ** included ** *** ********** at ** ********** ****. Users **** **** ******** one **********'* ******** ******** *** ** a ***** ***** **** server **** ********* *** (Nvidia ******* ** ******).

********** ****** ***** **** with ******** *********, ****** from*** (*** *** ****)**** *********** ** ************ ***** ***** *******.

********

******** ******* ****** *.*.*.** was **** ** *******. Nine ********* ** *** H4 ******** ******* **** used ** *******, ********* bullet, ****, ***, *** multi ******.

Comments (30)

John Honovich

IPVM | 10/30/19 03:18pm

********* *** ** ****** problems, **** ** * general ***** **** ** and ***** ***** *********, of ******, *** * think ** ***** **** a ******* ****** ** camera *********. *** **** people **** *** ** such *********, *** **** there **** ** * push *** ***-** / low ***** *******.

Michael Miller

10/30/19 04:01pm

** **** ********** ****** deployed ** ******* ***** systems ** **** ********* scenarios **** ******** ** cameras **** *** * am ************ ********* ** how **** ** *****.

John Honovich

IPVM | In reply to Michael Miller | 10/30/19 04:07pm

*******, *** ******** ********? That's * ****** ******** endorsement *** ** *** want ** ******* **** partner ****'* ******* **** ** ** with *********.

Michael Miller

In reply to John Honovich | 10/30/19 04:12pm

**, *** ******** ******* we **** ********* * Vape ******** ** * school ***** ** **** to ** ******** *** camera ******* *** ********. I **** *** * search ***** * **** alarm **** ** *** I *** **** ** right-click *** ** *** person ******* *** ******** and ******** ****** ****** me **** **** ******* the ******** *** ***** classroom **** **** ****. In * ****** **** 300 *******, ** *** impressive *** **** * was **** ** **** where **** **** ****.

Undisclosed #1

IPVMU Certified | In reply to Michael Miller | 10/30/19 05:40pm

** *** *** “******” with *** **** *****?

Michael Miller

In reply to Undisclosed #1 | 10/30/19 05:46pm

**... *) ****** ** not *** ****** ** people/vehicles ** * ***** needs ** ****** **** in *** ******** *) there *** ** ******* in *** ********* ** the ****** *****'* *** who ********* *** ***** only ****** ******* *** bathroom.

Undisclosed #1

IPVMU Certified | In reply to Michael Miller | 10/30/19 07:49pm

** ** **** ********** case, ** *** **** sequence ** ****** ****** tomorrow, ******* **** **** person ******’* ******** *** alarm **** *** *** before?

Michael Miller

In reply to Undisclosed #1 | 10/30/19 08:14pm

****** **** ******* ** do **** ********** ****** besides ****** *** ****** detected ** *** ***** event. ** ** * do ** ********** ****** over * ** **** time ***** *** ****** will **** *** **** person ** ******* **** the ***** ** ***** regardless ** *** **** alarm ****** ***** ***.

Avatar

Daniel Tyrrell

10/30/19 05:24pm

* **** **** ** tested ********* (********* *******, but *******, * *****?) they ******* *** ****** would **** *** ****** over ****. ** ** sees ****, *** ******, especially **** **** ***** (yes **** **** ** actually **** ******, ** that ** *** * face, ***).

** *** ***** **** is *********** **** ** Avigilon ** ****? * have **** ****** ** you ******* *** ****** tell *** ****** **** the ******* ** * puddle ** *** * face, ** **** *** face ** *** *** same ****** ** *** other ***.

**** ******** **** ** fixing ****?

Michael Miller

In reply to Daniel Tyrrell | 10/30/19 05:33pm

******** ** ********* **** Anyvision ** *** **** that **** *** ** an ********** ****** *** can ****** *** ******* that ***** **** ****** and **** ****. ***** time *** **** * result *** ****** ******** again *** *** **** stars *** **** ** the ****** ** ******* you're ******* *** *** better *** ******* ***.

** *** **** *** false ********* ** *** initial ****** *** *** more *** **** **** subject *** ****** *** results ***. * ***'* think **** *** ********* very **** ** **** test.

Undisclosed Manufacturer #3

In reply to Michael Miller | 10/30/19 06:09pm

************, ********* ** **** facial ***********. ******** ** classifying ****** *** ********. Appearance ****** ** ******* for ******** ******** ** classified ******* ****** **** facial ***********. ****'** **** different.

Undisclosed End User #2

10/30/19 06:01pm

**** *******, ****'* * passing ***** ** **** a ***** ********* **. red '*' ** ***** tests?

*** *******, *** ********** classification ** ****** "********* accurately", ***** ** ****, but * ****** ** giving *** ***** ***** on ~**% ********(*/** ****** shown *** ******* *** motorcycles) ** *** ********. Of ****** ***'** ******** looking ** * ****** data *** ** **** eval, *** **'* ***** to *** **** *** of *** ******.

Slava H

10/30/19 07:26pm

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

** ***** ** ***** to *** **** ********** test ** ******** **** Learning, *** ******* ***** on ****** ** ***.

**** ***** ** ***** to *** *** **** results **** ****** ** found. *** ******* ***** Classification - *** **** were ***** ** ******.

Michael Miller

In reply to Slava H | 10/30/19 07:44pm

***** **** ** **** with ******** **** ******* search ******* *** *** going ** ** ***% perfect. *** ** *** human **** *** ******* and *** ****** **** looks *** ***** *******. The **** *** **** a ****** ** ******* the **** ******** **** search **** **.

Undisclosed Integrator #6

In reply to Slava H | 10/31/19 01:50pm

* ** * **** fan ** ******** *** we *** ******** ********** search ***********. *******, ***** is *** ******* ***** some ************ **** ********* for ****** ******, ** is ********** *** ******** on ****** ** *** and *** ******* *** even ***** **** *** subjects *** ** * different ****. *** **** has ** **** ************ inaccurate ** *** ****** results, ***** *** **** misses. **** ******** *** absolutely *** ******** **** over ****.

Michael Miller

In reply to Undisclosed Integrator #6 | 10/31/19 02:01pm

*** *** ******* ***** the ******* ****** ******* or ***** *** **** your *******?

Undisclosed Integrator #6

In reply to Michael Miller | 11/01/19 10:20am

**** ********* *** ***** results *** *******.

Michael Miller

In reply to Undisclosed Integrator #6 | 11/01/19 01:41pm

* ***'* *** * agree **** *** **** it ***** ** *** starred *******. ****, ****-******** is *** *** ********* in *** ******, *** the ********** ******.

Slava H

10/30/19 07:45pm

*** *******, **'* **** interesting.. ****'* **** * want ** *** ** a ********** ********* (**** results).

Undisclosed Manufacturer #3

10/31/19 11:16am

* ******* **** ** not ********** ** *** scientific ***** ** *********** then ****** ******* **********. There *** *** **** variables ***** *** ****** the *********** ** ********.

** *** **** ***** scene, ** *** **** the ****** **** ** right ** ** ** down *** ******* ** an ********** ***** ** vastly ********* (** ***) when ** ***** ** analytics.

** ***** ** ** the ******* ** ***** analytics ** ********** *** probability ** ********* ** undesirable ************* ** *** field ** **** ** a ****** ** ***** to ** **** ********* in **** **** ********** and, ** *** **** of ********** ******, ********* reducing ************* **** **** with *** ********* ** false *********. **'* **** easier ** *********** ********* that ***** ** ***** rather **** *********** ********* you ***** ***** **** known ***** ******* ******** hours ** ********* *****.

Slava H

In reply to Undisclosed Manufacturer #3 | 10/31/19 05:05pm

* ***** ******** **** your ***** ********* ** sense ** ************* ***** being ************, *** ****'* not *** ***** ****.

**'* **** ****** ** investigate ********* **** ***** be ***** ****** **** investigate ********* *** ***** would **** ***** ***** without ******** ***** ** scrubbing *****.

***, *******. **** ***** would ** ** ****** out ****** ***** ** your ********, *** *** to **** ******. ***'* assume ***** ** *** criteria ** **** ****, and ** **** **** to *** ******. **** wise - * ***** like ** **** *** many ****** ** ****** vs *** **** ****** we *** *********, **** self ******** ******** **** ratio. **** ***** **/**** event ***** ** ******** model *** ***** ** worse ** ******* ****?

**** ******** **********: **'* easy ** *** *******/********* events ** ****** *** presented ** **. *** impossible ** **** ***** on **** ****** *** what ** ******* **** it. *** ***** *** know *** ****** *** Truck *** *** ******** should ******* ***, ***** you ****** ******* ***** of *****?

Undisclosed Integrator #4

10/31/19 11:42am

**** ** **** * see ** *** ****** of ****** ******* *** they’re ***** ** *** competition.

**’* * **** **** it **** ***** **** their ******* *** **’* expensive.

Michael Miller

In reply to Undisclosed Integrator #4 | 10/31/19 11:53am

*** *** ***'* **** to *** ******** ******* for ********* ******. **** I ******** ** *** expensive **** ** ********** Search ** ** ***** charge ****** ***** ******* that ******* ********** ******** and ******* ***** **** significant ***** ******** ** Avigilon.

John Honovich

IPVM | In reply to Michael Miller | 10/31/19 11:57am

********* ** ********. * significant **** ** *** market *** **** ********** to $** - $*** cameras, ****** **** ********* for ****. * ** not ******* ******* *** value ** ****, **** pointing *** **** **** a ***** ******** ***********, Avigilon ***** **** *********. And * ***'* **** about #*, *** * am **** **** *********** will ** ********** ****** "Hey ***** **** **** this ** {****** ******} and **** **** ** away *** ****."

Michael Miller

In reply to John Honovich | 10/31/19 12:01pm

******* ***** *** ********* to *** ******* ******** you *** ******* ** additional ******** ***** ******** more ******* *** ******** which **** *********** ***** compared ** ********.

John Honovich

IPVM | In reply to Michael Miller | 10/31/19 12:10pm

****, ******** ** ********* like ********. ****'* *** I ***** **'* ********. If *** *** ****** from *** "* *** spend **** ** ***** for * ******* ********" position, ******** ********** ****** may **** *********** *** if *** *** ****** from "* ***'* *** a ***** **** **** $100 *** *******" **** will.

Undisclosed Manufacturer #7

In reply to John Honovich | 10/31/19 11:37pm

******** *** **** ********* a **** ********* ******* and ******* * **** start. ** *** ****** months, ***** **** ** other ********* ************* (** least *) **** ******* learning ******** **** *** open, ****-*****, *** **** extremely ****. * *** months ***** **** *******, major *** ********* **** accept *** ********* *** metadata **** ***** ******* and ******\*****\******* ****** **** be **********. * ***** Avigilon **** ****** ** these ****** *************, ** least ***** ***** ** surmounting ********* ******** ( like ******** *.*** *** ONVIF).

* ****** *** ******** will ***** ** ** the ******** ***** ***** leveraged ******* * **** channel (************, ************ ****, integrators, *-********, **** ********) that **** ***'* **** (unless *** ***** ********'* radio *******?) **** ***** edge **********.

** **** **** ***** open **** ** ******** hit *** ******, *'* not **** *** ***** AI ******** ********** **** grow (*********, ********, ***) when *** **** ******* and\or ********* *** ** so **** *** ****** and ********** ** ***** dedicated ****** ********.

*** ****** ** ********* servers ** * **** strong ***** ***** ***-***** currently.

***** ***** **** ****** unbanned *** ************ *** take **** **** **, who *****??

John Honovich

IPVM | In reply to Undisclosed Manufacturer #7 | 10/31/19 11:42pm

** *** ****** ******, there **** ** ***** reputable ************* (** ***** 3) **** ******* ******** products **** *** ****, edge-based, *** **** ********* well. * *** ****** after **** *******, ***** VMS ********* **** ****** and ********* *** ******** from ***** ******* *** object\color\feature ****** **** ** mainstream.

*** **** ** *** a $* **** ** is ***** ** **** way ****** *** **** to ******? *** ** be *****, '******* ********' is ****** *****, ** you **** ****** *********, sure * *****, *** we ** **** ****, we *** ******* **********/********** search. **** * ***** is ***** ** **** a *** ****** ** be '**********'.

Undisclosed Integrator #5

In reply to Michael Miller | 10/31/19 12:18pm

** *** **** **** of ** **** ** get ******** ********. ***** not ** ** *** user, *** *** ******* I've ****** ** ***** trying ** ***** **** some ******** **** *** side **** * ** now *** ****. * can't **** ********, ** I'm *** ***** ****** enough ** ** * dealer, *** ***** ***** dealer **** ******** **** to **, ** * massive ******** **** ****. Either **** ** **** deem ** *** ******** and ***** *** **** to **. *'** *** into **** **** **** than **** ********. ** I'm *** ******** *** money, ** ***(*******) ***** to ** ** * hurry ** *** **** to ** **** *******.

** *** **** ** Avigilon ******* * **** expensive *******, * *** stuck ****** **** **** since * ***'* ****** a ******, ** ******* for ****.

Avatar

Hash Salehi

Motorola Solutions
In reply to Undisclosed Integrator #5 | 10/31/19 06:17pm

* ****** ******* ******** must ***** *********. **** of *** **** ******* came **** ********** **** as *** ********. * encourage *** ** ***** out **********.*******@********.****** ******* ** ****** a ****** ** *** are ***** **********.

* **** ******* ** speaking **** *** ***** the ********** ** *** past *** **** ** can ** ** ****** the ******.

Read this IPVM report for free.

This article is part of IPVM's 6,653 reports, 896 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