Genetec Kiwi Intrusion Detector Analytics Tested

By Rob Kilpatrick, Published Nov 27, 2018, 10:00am EST

Genetec has built Kiwi Security's Intrusion Detection analytics into Security Center, aiming to simplify deployment compared to separate camera based analytics.

IPVM Image

We tested Intrustion Detector to see how this compared against leaders from our IP Camera Analytics Shootout, examining:

  • Configuration and calibration requirements
  • Outdoor false alert performance
  • Snow/rain false alert filtering
  • Indoor false alert performance
  • Detection range/PPF requirements
  • CPU/GPU load impact

Positive - ****** ********

******* **** ********* ******** offered ***** ********* ***********, avoiding ***** ****** **** common ******* **** ** swaying *****, ***** *******, and ******* ********, *** was *** ** *** analytics ****** ** ***** alerts **** **** *** snow ******** **** *** dome ** *** ******. The **** ******* ****** of ***** ****** *** major ***** *******, **** as ****** ******* ***/** in * ****.

Negative - ****** ******* *****

*******, ********* ******** ***** was ***** *** **** complex ** ********* ** have ******, *** ****** over-adjusted ** ** *** sensitive ** ***********, ******* false ****** ** **** small ******* ** ******* to ***** ** ****** in *** **** ***** of ****.

Notable - *** ****

************, *** **** *** camera *** ~*% ** our *****. ** **** small ******* **** ****** may ** **********, *** in **** **-** ****** systems, ***** *** **** to ******* ********* ******** to * ********* ******.

*******/************

******* **** ********* *** an **** ** $*** USD *** *******. ********* do *** ******* * separate ******** *******, **** enabling *** ********** ******** role ** ******** ****** (though **** *** ******* an ********** ****** ********* on ****, *** *****).

Complex ***** *******

*** *** *********, ******** configuring ********* ******** *** simply ** *********** **********. In *** *****, ** were ****** ** ********* Kiwi ********* ** * basic ***** ** ****** following *************, ********* ******** conversations **** ******* ** properly ********** *** ***** process *** *** ******** options ********. *** *********** who **** *** ******** this ********/********, *** ******** may ****** ***** **** right ** ******* ******** return ***** *** ***************.

***** **** ******* *** objects, *** **** *** one ***, *** ****** detection ***** **** ** cover ******** **** ***** 4 ****** ** ******* ranges (** *********** ** Kiwi). **** ******** * significant ****** ** ***** and ***** *** *** be ****** ******** ** be **** ** ***** sensitive, ********** ***** ****** or ******* **********.

*******, **** **** **** testing/retesting **** *** **** to ** ********* ****. Users *** ****** * region ** ***** ** loop *** *******, ** a ****** **** *** perform * **** **** and *** ***** ** themselves *** *************, ****** many ********* ***** ******* live **** *******, ********* with *** ******. *** video ***** ******** *** process:

Strong ***** ***** ********* ***** ******* *************

***** ********** ***** **** as *** ****'* *************** (~4 ****** *** ***** subject) *** ***** **** (2 *** *******), **** managed ** ***** ***** alerts ** ****** ******* which ********* ***** *********, including ***** ******* ** the ****, **** *****.

no-alerts-on-swaying-brush-foliage

******* ****** ****** ** false ******, ***** ******* (e.g. *** ****** *****), were ********** ******* ****** testing.

no-alerts-on-small-animals

False ****** ******* ** ****/****

***** **** ********* ******** from *********** ******* ** false ****** *** ** rain *** ****, **** includes ** ******** ****** to ****** ***** ******. In *** *****, ******** this ****** ******* ****** from **** *** **** both ******* *** ******** on/down *** ****.

no-alerts-on-snow-rain

** *** *****, ***** the ****/**** ****** *** not **** *** ****** on ********* ** *** scene, **** *** ********* still ******** ********* *** subject *****.

snow-rain-filter-does-not-impact-detection

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

*******, ********* ******** ******** from ***** ****** ** major ***** *******, **** as *** ****** ***** turned ** *****. ** did *** ***** ** other *******, **** ** shadows ** *********** ****** in ****** *******.

false-alerts-on-drastic-light-changes

Long ********* *****

**** ************ ******** ******** at ~*-* *** **** day *** *****, **** of *** ****** *** requirements ** *** ********* we **** ****** (*** our** ****** ********* ********. **** ******* ** the ********* ******* ******/*** widths ** ******* ******:

IPVM Image

Increased ***** ****** ** **** *****

********* ******* ** ****** ranges ****** ** ******** false ******, ****** ** small ******* (**** ** vehicles ***** *****) ** the **** ** *** field ** ****. ********* Detector *** ******* ** detecting ******* ************ ** ~220' **** *** ******* (~490' **** *****), ****** this ********* ***** ********** moderately. ***** ********** ***** be ******* ** ******** configuration ******* (*.*. ********* dwell **** ** ******* distance ********), *** ***** options *** *** ********** remove ***** ******.

long-range-detection-increases-false-alerts

CPU *****/******** ************

********* ******** *** * significant ****** ** *** and *** ***** ** our *****, ********** *** by ~*% *** ****** and *** ** ~*% when ********* ** * lower ********** *** ****** (recommended ** ****). **** analyzing ** * ***** stream (*********** **** ** specific ************ ** ****), increases *** **** **********, with *** ***** ******* by **** **% *** GPU ** ~*%.

IPVM Image

******* ** ***** *******, those **** ******** ***** than ***** (of ******* *** **** to **** ** ********** analytics ********** ** * separate ******. ******* ****** that ******** ******** ** Kiwi ********* **** ******* less *** **** ** more ********** ** ********* to *** ***, ****** no ******** *** *** given.

Drawbacks ******** ** ****** ********* (********, ****, *****, *****, *********, ***.)

**** *** *** *********** drawbacks ******** ** *** the ****** ********* ** have ******: ***** ********** and *** ****.

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

****'* ******* ***** ** by *** *** ******* disadvantage ****** ****** ********* we **** ******. *** the *********, ****** ************* may ****** *** ** possible.

**** *** ***** *** have **** *******, ******** configuring *** **** ******* Intrusion ******** ***** ************* more **** **** ***** other ******** ** *** shootout, ****** **-** ******* per ******* *** ****** more *** ************* *****, including ********* ***** ****, scene *****, ***** ****, and ***** ******** ********.

** ********, *** ** the ********* ********* * longer *********** ******* **** most, ***** *** *** Axis ********* ********, ******* closer ** ~* ******* per ******. ******** ******** even **** ****, **** users ********* ******* **** to ****** * ****, and *** ****** ****-******** the *****.

*** ****

************, ** *** *** the ******** ******* (~** cameras *** *****), ***** must ****** ******* ** additional ****** *** **** processing, *** ********* **** using ****** *********.

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

******** ** ****** ********* from ******* ********* ********, **** ******** **** in ***** ** ***** alert ********** *** ********* distance.

  • ***** ***** **********:********* ******** ******* ***** alerts ********* ** *** top ********** ** *** tests, ******** *** *****, rejecting ****** **** *******, small *******, *** ** shadows, ** **** ** snow *** ****, ***** triggered ***** ***** ******** in *** ********. ****** such ** ****, *****, Hikvision, ***., *** ******** from ***** ****** ** at ***** *** ** these ***** *******.
  • ********* ********:**** ******** ***** *** for ******** ********* **** most ***** ********* ******, day *** *****, **** only **** ********* ******** performing ********* (~* ***). Others **** ** ******** and ***** ********* ******** 8-10 ***, **** *****, Hikvision, *** ******* ********* higher *** ******, ~** or ******.

Software ******* ****

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

  • ******* ******** ****: *.* SR3 (*.*.***.**)

Comments (17)

As a Genetec dealer I feel better than we were not the only ones who struggled to get it to do anything. :)

John and team, how does IPVM measure CPU/GPU usage? 

I ask as I'm curious how the percentages you listed above play out over a 24 hour period with different degrees of scene complexity or busy scenes with lots of classified objects.  Is there a plus or minus X% usage range?  Thanks for including the CPU/GPU usage; I think this will be an important metric for evaluating such systems going forward.  

That's a good question. It's sort of a spot-check, but it's checked multiple times, typically. We don't just open up the PC and measure once and call it done. We tend to check throughout testing, in different conditions, etc., and give a realistic average if it doesn't vary too much from scene to scene, or specifically call out where it's higher or lower. In this test, it was fairly consistent throughout, but a +/- might be useful to add. We'll take a look at results on this and keep it in mind for future tests.

This post is obviously fairly old, but this is the latest that I could find on Kiwi.

This specific post sort of addresses my question, but I wanted to try and get a bit more into the specifics here.

Can you share the exact specifications of the server that were used here?

I ask because saying that each Kiwi instance increases CPU load by 5% isn't necessarily a great measure -- the CPU load on a Core i3 will be dramatically different than the CPU load on a Xeon Platinum (as an extreme example).

I plan to test this myself soon enough, but getting an idea would be helpful in advance.

Thanks!

We offloaded Kiwi processing to our test laptop while the Genetec server itself was running on an Intel NUC. The test laptop at the time was a quad-core i5, 1.6 GHz. Most of our machines now are higher spec so I would expect a little bit lower CPU usage, but we have not tested it since this test was released.

Hi Ethan,

First of all, thank you for your comment in this regard. My name is Fabiola and I am the Commercial Lead for Video Analytics at Genetec. To provide additional information on the topic I would like to mention that resource requirements differ based on the camera resolution and the video analytics scenario you would like to run. To make it easy to find the right hardware configuration, we provide a dedicated tool, which allows you to calculate how many analytics you can run on a given server. If you have access to the Genetec Technical Assistance or Channel Partners portal, you can find the Hardware Calculator under the point “KiwiVision Hardware Calculator”. This is the easiest way to scope a project and figure out what hardware is required for a given configuration. Also, in the Security Center 5.9.3 release, we added Analyzer management, which automatically balances the load between servers as well as CPU & GPU."

If you have any questions about resource requirements, do not hesitate to reach out to our support teams as well who are happy to help.

Genetec partners have access to a Kiwi analytics CPU calculator in the Genetec portal. Genetec has also just released load-balancing for analytics and further NVIDIA GPU offloading in the latest release a few days ago.

False Alarms Avoided on Rain/Snow

One thing that you forgotten to mention is that when the camera is dirty or covered with snow like that it won't false alarm but probably won't detect anything either.

Hi Rob,

I’m the product marketing manager for video analytics at Genetec. First off, I want to thank you for the great review. We’re working on making the Intrusion Detector setup easier for system integrators. In the meantime, we do offer technical certification training as well as field engineering services.

Stand by for upcoming updates that will ease the deployment and configuration of our unified video analytics modules.

Hello Laurent,

Can you tell me what training class to take to learn about Kiwi Vision?

When I looked, 3/31/2017, for training on it the course outlined to me was SC-KV-001.  Noting OTC-001 is required for the KV-001.Training for Kiwi

Do you see a major advantage to server side analytics? Kiwi is obviously server side and seems to cope very well with real world outdoor scenarios compared to the camera side on Bosch/Axis etc in your shootout test. I'm thinking of a perimeter security use case here which is what Kiwi is primarily targeting where the rain/snow filter seems to be a huge advantage Kiwi have in real world applications. I imagine it would be very difficult for others to do this with any level of success on a camera due to resource limitations. 

Cheaper cameras seems like one big advantage.

I was wondering if any of the folks that conducted this testing (or any VMS manufacturer for that matter) are certified in those particular products?  I do not mean this as a negative comment but just curious as VMS' tend to require some more training etc.

We generally talk directly with manufacturer product management / engineering when doing tests. Regular 'certification' training is generally quite basic and not useful to us. 

Good day, How would this stack up against the Avigilon and Agent VI systems as I think those will be the major competitors for VMS side analytics.

This thread is 2 years old. Any updates on the comparison between Avigilon and Kiwi?

Read this IPVM report for free.

This article is part of IPVM's 6,817 reports, 914 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