Avigilon "Self-Learning" Analytics And VMD Tested

By IPVM Team, Published Jul 23, 2018, 11:55am EDT

Avigilon is a popular choice for video analytics, offering "self-learning" analytics built into their cameras as well as appliances.

We tested their self-learning analytics indoors and out on Avigilon's H4 dome cameras, as well as their standard camera VMD, to see how it performs in real applications.

Inside, we look at:

  • Self-learning vs. calibration
  • False alert performance
  • Detection range
  • Analytics compared to camera VMD

Note that this is the first of our updated tests of Avigilon's analytic offerings. Tests of Appearance Search and Unusual Motion Detection will follow in upcoming reports.

This report continues our recent series of updated analytics tests. Readers should also see our tests of Axis Guard SuiteAxis VMD4Bosch IVADahua Intrusion AnalyticsHanwha Wisenet X Analytics, and Hikvision Intrusion Analytics.

*******

** *** *****, ********'* ****-******** analytics worked **** **** **** properly **********, **** *** ** no ***** ******.

  • ** ***** ****** **** common *******: ******* *****, ***** animals, ***** *******, *** shadows did *** ***** ***** alerts ****** *******, ******* or ***.
  • ** ***** ****** **** rain: ******** ******* ****** from *** ********* ** rain, *-* ****** ******* from ***** ** *****, one ** *** **** common ******* ** ***** alarms ****** **** *********** analytics.
  • ***** *** ******** ** well *** ******: ** *** *******, ******** ******** ******** human ******** ** ~*-** PPF, ************* ****** **** most ****** *********, **** as ******, *********, *** Dahua (**-** *** ** average), *** ******* ** slightly ***** **** **** Guard ***** (~**-** ***) and *****'* *********** ** (~10 ***).

******** *** **** *** shortcomings ******:

  • ********* ******** ********* ***** * lux: ********* ******** ** ******** cameras ******* *********** ** low ***** ****** (lux), **** ~***' ******** at ***** ** *** 5MP *** ** ****** (7/10 ***, ************) ** ~55' (~**/** ***). ********* were **** ******* ** the *** *****, ******** to ~***', ********** *** required **** ~* ** 10.
  • *********** ***** ******** *** best ***********: ***** **** *** ****** "self ********", ********'* ********* require ****** ******* ** ***** the ***** (****** ** vehicles). **** ***** ** static ****** **** ****** or ******** *** *** normally *******, ***** **** see **** *********** ** manually ******* ** ******* through *** ***.
  • ******* *********** ****** ******** period:*********** ** ******* ** these ******* *** *** present ** *** ***** for *** ****** ** learn, **** ****** ** small ******* ****** ** *** testing ****** **** ******.

************

******** ****-******** ********* *** available ** **** **********'* *** ****** *******, ********* ***,****, ******, ***,*** ******(**/** ****), ***, *** thermal *******. ********* *** not ********* ** *** SL ******,** **** ****, ******* ********** ** *********** (****** **** *** ********** ***** ******** ********* multi ******).

Calibration / ****-******** *******

***** ********'* ********* *** self ********, **** ***** require ***** *********** ** order ** *********** ******* desired ******* (****** *** vehicles) *** ******** ******, such ** ***** *******. According ** ********, **** amounts ** ***** *** *********** ** *******, .

******** ************* *** *** require *** ****** *******/******* through *** ***** ** many ***** ***** ***** is ******* ********. *******, in **** ****** ******, such ** **********, **** fields, "** ***'* ****" areas, ***., ***** ******** are ****** ******* ***** normal *************, ******** *** take *****, **** *********** reduced ** *** ********.

*** *******, ***** ** calibration, *** ******* ******* a ********* *** * vehicle. ***** **** ******* calibration, ***** ******* **** reduced ** **********.

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

******** ********* ******* *********** was ************ ***** ********** testing, ******** *** ********* of ******* ***** / foliage ****** **** *** weeks ** *******, **** during **** **** ****, below.

********'* ********* **** ******* alarms **** ***** *******, another ****** ******* ** many *********** *********. **** that **** ******* ********* were ***** **** ** properly ********* (*** *****).

*******, ***** **** ***** analytics ********* ********** ** rain/droplets ** ****** *****, Avigilon *** *** ***** during *** ******* ** rain **** *** ****** of *******.

Indoor ***********

*******, *** ********* ******* false ****** **** ******* such ** ******* ***** changes, **** *****.

***** *******, **** ** light ******* ** ***** rooms *** ******** ******* on/off **** **** *******.

Vs. ***** ****** *********

** *** *****, ****-******** analytics ****** ************ ********'* standard ****** ***, ***** Avigilon ***** "***** ****** Detection", ***** ******* ** practically *** ******** ** the *****, ********* ******* cars, ***** *****/*******, ***. Sensitivity *** ********* *** be ******** ** ****** these ******, *** **** caused * ***** ******* to ** ********** ****** walking ******* *** ***** (shown *****).

Detection ******** ******* ***** * ***

******** ********* ************ ******* on *** ******* ** ~7-11 *** ****** *** day *** ********* *** drastically ******* ****** ***** testing ** ~**-** *** which ******** ** *** following ***** ** **** widths / *********:

  • *.**-***-***-** (***/*:*): ~***' ****/~***' Distance ** ******** *** ~250' ****/~***' ******** ** night
  • *.*-***-***-** (***/*:*): ~***' ****/~***' Distance ** ******** *** ~90' ****/~**' ******** ** night
  • *.*-***-***-** (**/**:*): ~***' ****/~***' Distance ** ******** *** ~95' ****/~**' ******** ** night

********** **** ******* **** using **** ****** ** view, ****** ****** ** view, ******** ** ** and ***, **** ***** filtering ** *** ***, sensitivity *********, ***. ******** changes *** ** ********* effect ** **** ********* range.

** ******** ***** ********** with ******** ******** *********** who ********* **** **** either **** ********* ** brighter ****** (*.*. *** parking ****) ** ***** supplementary ********, **** ** external ** ************ ** additional ******* *****.

**** ******** ** ******** was *** ******* ** other ******* ******, ***** saw **** ****** ** moderate ********* ** ********, but *** ** ****** as ** **** ****. For *******, ****' ***** Suite ********* **** **** to ****** ** ************* the **** *** ****** *** *** night (************* **-** ***), as *** ***** *** (~10 ***). 

Only ********* **** ***

********'* ********* *** ********* when ***** ***** *** Avigilon ******* ****** ***, which ** ******** *** event *************. ***** ** no *** ********* *** event ***** *** *** camera ** ** ****** in **** ***** ****** analytics. ************, ******** **** not ********** ********* ***** analytics **** *** ***** party *****, ****** **** point *** **** **** use *** ***** ***** framework, ***** ** ********* to ****** ** ******* these ******** ******. *******, ACC ***** ***** ** required *** ************* ** ***** cases, ****** ******** ********* and ***** ***** *** an ******** ***********.

Versions ****

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

  • *.*-***-***-**: *.**.*.**(*****)
  • *.**-***-***-**: *.**.*.**(*****)
  • ******** ******* ******: *.*.*.*

Comments (27)

This report is very consistent with our findings.

At the moment Avigilon has the best Analytics system our there.

The only problems are price and 3rd party integration...

Great review really.It would be interesting also (i know it sounds funny)to "test" how the camera performs under spider web.It happens so frequently if the site is not supervised by 24/7 guard that some insect especially spiders can build their web on camera's glass.Especially in winter time and that might stay for hours untill somebody cleans it. That's the most challenging thing for analitics especially at night time with IR on and i've never seen a camera without millions of false alarms under that condition

From our own experience in our Avigilon demosite, I can confirm that spiders/ spider webs in combination with IR and wind did not have any effect, while it was still capable to detect persons and vehicles. The camera used was a 2.0C-H4A-25G-B01-IR. 

In case I see that there is high risk of spiders and insects, I always recommend to my customers to install separate IR illumination 50 cm to 1 m away from the camera, and of course use box camera or turn of IR on bullet camera

Well i don't know if it's because here we're in the countryside but EVERY outside installation sooner or later has problems with insects.I'd rather choose box camera (also for better performance compared to compact) but that's a lost battle (aesthetic,price,smaller range of products,ecc. ecc.)Funny as it sounds i personally spray every camera's body with anti insects spray (the same you use to keep them out of the house)...it works for a while then customer must take care...

Luca what a great idea.  We have that problem and I never though of that.  In fact, I think they make a sticky paper like fly paper that you could put on the camera that might last even longer than the spray given weather issues.  Hmmmm

Yeah that could be an idea...never tried it but i guess that sun and rain make that paper less and less sticky day after day...spray works fine the point is you can't put it on the glass in front of the lens but i think that the smell of that spray all over the camera is just enough to take the animals quite far away...the idea came from my older collegue who is a specialist in automatic gates a field that has big issues with ants,lizards and even small serpentines in the control panel that is usually placed close to ground level...the basic point is someone has to take care of a periodic maintenance even in terms of cleaning when you place electronics in the outside

Solar powered bug zapper, hat tip U1.

This is absolutely correct for every analytics installation including Avigilon.

They all have problems with insects and spider webs just plain dust.

Use box cameras with the IR installed 50cm below the camera and you'll eliminate most if not all of these problems.

 

Why use a Box camera when you can just turn off the IR on the bullet and dome cameras?

Like i already wrote the box camera/non box camera topic is not even considerable...i am in this market since 20 years i started when cameras where only box black and white CVBS 300 TV lines. I'd personally choose box camera on every site since i exactly know all the benefits of using a box camera with separate IR the fact is….box cameras have been outwon by compact cameras.Maybe in some specific installations end costumer would allow the isntallation of a box camera but box cameras are ,unfortunately,stone age

You'd still have to use a separate IR light then and he's saying put it under the camera.  You wouldn't just turn off IR and not add light another way or you won't achieve the picture that drove you to use the IR functionality to begin with.

Sorry, I thought the adding a separate IR unit was a given. 

Avigilon analytics appears to be reliable and the fact that it in compasses calibration/self learning features, this indeed is great tool for reducing false alarms. Also the outdoor and indoor performance in the above video tests are quite impressive as they are stable.

Thank you for this interesting report. A couple of questions:

- Did you compare CPU load of standard motion detector and this algorithm (during the self learning process and after it)?

- Does it require GPU or not?

The questions are not actual. I've missed that this algoritm works on-board of the camera.

Great achievement, Avigilon!

Respect!

Such a good quality of tracker means very effective archive searching functions.

We need to check if integration with 3rd party software allows to transmit this tracks to server for further storing and analyzing during archive investigation.

Our typical problem is that server analytics is too havy for processing - means too expensive. If now cameras can provide such quality tracks, than problem solved!

I would like to see Avigilon amend its classification to "human", "vehicle" or "other" or something of the like. I run into a number of applications where alerting on "threats to infrastructure" are commonplace. Knowing that a family of rats, or groundhogs, which can damage property are also interesting to some customers.

Snakes and squirrels...

Any idea about avigilon camera (/w self-learning capabilities) pricing ranges comparing to legacy VMD cameras?

What do you consider a 'legacy VMD camera'? Old cameras from a few years ago? New Dahua and Hikvision cameras with poorly performing analytics or?

VMD which is based on pixel changes without any "learning" capabilities. once you can find in almost every surveillance camera.

Avigilon H4 outdoor domes run from about $800-1,200, depending on the resolution. That's estimated street price, not dealer price. 

I'd say that's fairly comparable to VMD only outdoor domes from years past, but on the high end of pricing in 2018. For example, you can get a Hanwha Wisenet Q 4MP dome with motorized zoom and no analytics for ~$300-350. But, if you look at the Wisenet X series, which includes analytics, pricing is more comparable, in the $700-800 range.

Hikvision's 4 series domes, their higher end models, are notably less expensive (~$550 for Darkfighter 1080p and up) and include analytics (though performance is questionable in many scenes). 

Thank you Ethan, very helpfull?

Sorry , meant very helpfull!

Interesting what ASIC they use, which allows deep learning classification on edge.

Can somebody please clarify the distances. Is it Yards, Feet, Meters?

  • 8.0-H4A-D01-IR (4K/16:9): ~340' HFOV/~200' Distance in daylight and ~95' HFOV/~55' Distance at night

Thanks

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