Axis Video Analytics Are Weak

By Ethan Ace, Published Oct 05, 2015, 12:00am EDT

For more than a decade, video analytics has frustrated and disappointed users.

Now, Axis has released their own "series of robust video analytics applications" that they call Guard Suite [link no longer available]. Unfortunately, IPVM's testing shows that Axis' video analytics are weak. We tested both the Q1615 and the P1428E with the motion and fence applications ($149 MSRP each but free 30-day trial applications are available here [link no longer available]).

******** ** ****' ******** effort,*** *, **** ***** ***** sources **** **** ******* or **********. ** *** tests, ***** ***** *** not ******** ******* ******, birds, ** **** ** *******, occasional ******** ** *** 3. 

** **** ***, ******** size, **********-****, **** *** weather, ********** ******** ****** slowly **** ***** ***** should work ********** *** ********.

*********

**** ***** ***** ** dumb.

***** ******* *** ** the ******, *** **** Guard ***** ********* ** reliably ****** ***-***** **********. **** modest **** ***** **********, false ********* *** ***** negatives ****** *********** **** Guard ***** ********** ** recognize **** ** * real ***** ** ******, wires, *******, ***. ***** problems *** *** **** severe ************ (******)******** ******* (******).

*********** *** **** ** intelligence is * ********** ***** process **** ***** ******** controls ** ******* **** might **** ********** *** this, ****** ********, ***** largely ******* **** *************, and ***** ******* **** provides ************* ***** *** calibration ** ******* ***********.

Market ****** / ***************

** *** *** **** Guard ***** ****** *** use ** ** ****** indoor ****** ** ****** in ** ***** *** week ** ******* ** the ***** ******** *** plan ** *** **. Despite ****' ****** ******, Guard ***** ** *** 'robust'.

** ****** *********** ***** this ** ** ********** by *** **** *********** and *** ********* ** overcome **. ** ** not *** **** ***** widely ****, ******* **** to ******** ** ***** on *** ***** ******** apps.

Performance ******

** **** ******* ** ***** down *** *** ****** impacting **** ***** ***** performance, ********* ***** ***** issues, ****** ***********, *** more, ** **** *****:


***** *********** 

****' *********, **** *** 3 ****** **, ****** from ******** ***** *********** from *****, ********, *******, and ***** **********. ** the ***** *****, *** brush ** *** ***** side ** *** *** is **** ** ******* by *** ******, ******* calibrating *** **** *** perspective. **** ******** ** far ** ** ************ cause ***** *********** ** the ******** **** ** the *****. 

***** *********, **** ** Avigilon *** ******* *** not ********** ***** ******** activations.

****** *********** 

** *** *******, ******** crossing * ******** ***** Axis ***** ***** **** often *** ********. ** the ***** *****, *** camera ****** **** *** detect *** ******* ** an ****** ***** ** is **** *** ****. Avigilon's *** ****** *** 100% ******** ***** **** same *****.

** ******* ******, **** was **** *****, **** Axis ******* ******* **** than **% ** *** time, ***** ******** *** successful >**%. ** *** example *****, ******* *** subject ***** **** ** an ****** ***** ** running, *** ****** ***** to ****** *** **** he ****** ******.

**** **** ** ***** FOVs, **** **** **** to ****** *** ******* as * ***** ******, tripwire *********** ****** **** not ** ****** ** *** examples *****. *******, **** still *** ******* ********* running ********, ******* **** about **% ** *** time.

** *****, ****** ********** at ****** ****** **** frequent, ** ****, **** with *** ******* ** well ***** ****** ******** areas ** *** *****, reviewed ** **** *****:

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

***** * ~**° ********** field ** ****, ***** Suite ********* **** ******** at * ******* ***** of ***** ***' ***** *** 1080p ***** ****** *** day. **** ** ***** **% less ******** **** *** Avigilon *** ******, ***** covered ~***' ****.

**** **** **** ** the ******* ***** ******** detection *** ******** (>**%). Beyond ****, ******** ** normal ******* ****** *** **** frequently ******, ****** ********* spotty.

Setup *** ***********

**** ***** ***** ********** perspective *********** *** ***** ** their ******** ************, **** as *** *, ******** to ****** ********* ******** and ****** **** ** measuring ******* ** *** FOV.

**** ******** ******* ******** on **** *******, **** instructing ***** ** *** measurements "** ********* ********* from *** ******."

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

******** ** *****, ***** also ******** *********** (******** does ***), ***** *** several *** *********** ** this *******:

  • ** ******* ****/***** ***** window:****' *********** ************ *** performed ** * ***** window, **** ** ****** to ********* **** ** enlarge *** ****** ** more ********** ******* ******** as ** ***** (**** A ** *** ***** below). **** *** **** it ********* ** ****** measurements ** *** ***** top *** ****** ** an ******, ******** ********.
  • ***** ********* ** ********:**** **** * **** to ******* *******, **** a ***** *** ** either *** ***** ********* size, ***** *** ******* the **** ** *** object ***** ********. ***** uses * ******* * shape (**** * *****) to ******* *** ****** scroll ***** ********** ** size, ****** ** ****** to ****** ** *** top *** ****** ** an ******.
  • ** ******** ***********:**** ******** **** ****** of *******, *** ******** calibration ***** ** ***** (and *****'* ****** *** releases). (**** * ** image *****)
  • ** ******* ****:***** ******** * ******* marker ***** ** **** to ****** ****** *********** of *** ******, ** well ** ******* ******* in *** ***. **** does ***.

************, **** ******** ** visual ************ ***** ** check *** ******** ** measurements, **** ** ****** or ********, ** ****** measurements *** ********. ***** and ***** **** ******* this *******.

*******, **** ******** ** maximum **** ******** ** vehicle **. ***** **************. All ******* *** ********, regardless ** ****/*****.

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

** *** ***** ***** we *********** *** **** calibration ******* *** *** issues ***** *****:

VMS ***********

*********, **** ***** ***** is ******* ** ****** integration *** *** ****** or ***** ***** ****. It ** ****** **** integrations **** ** ******** in *** **** *** months, **** **** *******:

"**** ***** ***** **** be ********* ** **** ****** ******* *** ******* *** ********* Genetec *** ******* ******* through **** *******’ ***** stream, ***** ********* ******* will ******* ****** ***********. Support ** ******* ****., Exacq ************ *** ***** is ******* ** ** available ** *** *** of ****."

** ** ***, ***** may ******* ***** ****** by ********** **** ** the ***** *** ****** in *** ******, ** seen ** *** *** of *** ****** *****. Users *** *** * motion **** ** ********* motion **** ** *** area ***** **** **** appears, *********** ******** ** the ******** ****:

***** *** **** **** TCP ****** ************* **** the ****** ** ***** which ****** ****. **** is ************ ** *** video ***** ***** ***** the *********** ** ***** Suite **** *****, **** event **** ********** *** ********** overlaid **** **** *****.

Comments (9)

Use digital barriers on a Axis - UK iLids approved. IMHO it just works :)

Agree: 2
Disagree: 1
Informative: 2
Unhelpful
Funny

The analytics developed by Axis was informed to target a low-end segment of VCA, although a step up from their VMD and simple cross-line it was not to become a high-end VCA (yet at least). The argument was not to compromise their open platform by competing with the 3rd party higher-end VCA solutions that can run on their cameras.

What I'm a little bit surprised about is the price, I had anticipated that they were going to offer this for a much lower cost, more in line with their previous Cross Line Detection.

Agree: 1
Disagree
Informative
Unhelpful
Funny

I think it is dangerous to offer analytics for a 'low-end segment of VCA'. It is very hard to determine how poorly the analytics will work in different scenarios, which makes it likely that they are deployed in conditions they cannot handle well, which makes customers frustrated and causes integrators to lose lots of money troubleshooting / trying to optimize.

Axis lead marketing claims for Guard Suite is "Robust and reliable indoor and outdoor detection" that "Protect and detect around the clock". If they really wanted it to be 'low-end', then they should have avoided such claims. As is, they have set expectations that they cannot meet.

Agree: 2
Disagree
Informative
Unhelpful
Funny

Yes, unfortunately marketing talk will almost always over-sell products.

As for targeting the low-end, that's what they did with the Cross Line (in my opinion). It seemed to perform rather well compared to its competitors in the tests you performed. But I guess their marketing language wasn't as bold then.

http://ipvm.com/report/analytics-rain-axis-bosch-sony

http://ipvm.com/report/camera-analytics-tested

I would personally be extremely reluctant to use the Axis Cross Line in any outdoor environment since the consequences of missed alarms and a high false alarm rate would be devastating to our costs and brand value.

It will however be interesting to see how well the new Axis analytics perform. Hopefully it is a step up from the old Cross Line and you seemed to think that it at least reduced the false-alarm rate somewhat compared to VMD 3.0. Perhaps it can perform good enough in ideal, sterile environments (as you mentioned in the test).

If you have time for it, it would be interesting to see how the latest Bosch-analytics compares with the new Axis as the latest Bosch should be much better than the old.

Similarly to the Bosch analytics (as well as just about any other VCA on the market that is actively developed), I think we'll see improved performance over time with the new Axis solution, the question is when (or if) it will be a good enough to use in non-ideal scenarios as well and whether it will be enough to be competitive towards the other camera manufacturers. Only time will tell...

Agree
Disagree
Informative
Unhelpful
Funny

Axis Crossline definitely had more false alert issues than you see in Guard Suite. I recall headlights, shadows, and blowing leaves being bigger issues than even what you see here. Guard Suite is at least far less susceptible to the headlights/shadows.

That being said, Crossline also didn't seem to have the same rate of missed objects that Guard Suite does. That's likely because it detected everything as an object, like debris/headlights/shadows, so it was less choosy and less likely to drop an object once detected.

Agree
Disagree
Informative: 1
Unhelpful
Funny

I would love to see an on-board analytic shoot between Pelco, Axis, Samsung, Bosch etc..

Agree
Disagree
Informative
Unhelpful
Funny

The placement of bars for calibration in you tests (see the "Axis Perspective Calibration" Figure) is certainly wrong because bars should be normal to the ground plane. Otherwise the data essential for calibration will be inaccurate. This is best seen with bars 1 and 2 drawn over the two cars. The bars should go along the front door vertical edge (which is not at all "vertical" in the picture).

I don't know if Axis has made a note of this fact in its manual but no calibration is possible with bars drawn at random.

Agree
Disagree
Informative: 1
Unhelpful
Funny

The bars are not drawn at random. The bars you see on the cars were drawn along the door line from the roof to the ground as you point out. However, Axis snaps the lines to vertical after they're drawn, but they remain the same length. I measured the cars those lines were drawn on, as well, to be accurate.

The cars themselves don't appear level due to the angle of view. The FOV overall is level, though, as you can see the poles in the background are vertical.

Also, this scene was later recalibrated when I moved cameras back to this point for further testing.

For reference, Axis gives little guidance as to what is right/wrong in this calibration, nor any best practice in their manual.

Agree
Disagree
Informative: 1
Unhelpful
Funny

Thanks for your clarification. Really strange that Axis "snaps" the bars to vertical... No way to double check the bars had been drawn correctly.

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

This article is part of IPVM's 6,938 reports, 926 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