This is the latest in our series of updated analytics testing. Next up are Avigilon's H4 analytics (camera based, with Appearance Search and UMD later) and Axis Perimeter Defender.
Readers should also see our others in this series:
Axis Guard Suite Video Analytics Tested
** ****, ** ************' ***** ***** ********* "****", **** ****** ********** *** ***** alerts ******. *** ***** ****** * years *** * *.* ******* *******, has **** ********?
** ****** ****** *****, ***** *****, *** Loitering ***** ** **** ***, ***** two *******:
- *****-***(*** ****** ****)
- *****-*** ** **(***** ****** ****)
- ************* *** ***********
- ***** ***** ***********
- ********* *****
- ********** ****** ****
- ****** *************
******* ****** **** *** *** ***** updated ****** ********* *****, ********* **** ****, ***** *** ***** ********* *** ******+ VMD, ****** ******* * *********, ********* ********* *********, *** ***** ********* *********.
Guard ***** *********** *******
*** ***** ***** ************ (******, *****, Loitering) ******** **** ***** ****** ****** by *****, *******, *** **** ********, **** similar *********** ** **** ** ***** issues.
************, ********* ******** ** *** ************ was *********** ******* **** ********** ******* object ******** *** ** **** ** recommend ********* ***** ***** **** *** FOV ******* ** ********** ** *********** them *** ******** *******.
*******, ** *** **** ****, ********* of ****** **** ***** ******* *** solid due ** *** ******** *********** ***********, with **** *** ****** (* *** *-* days) ****** ** ******* ** * week ** *******.
Detection *********** *********
** ******** ** ***** ******* ********, we ***** ***** *** ******** ******** to **** ***** ***** *********** ** our *****:
****** *****
****** ***** ******** ******** ******** ** ~10-12 ***, ~***' ********/***' **** ** the ***** *** **' ********/~***' **** in *** *****, **** ** ***** negatives/missed ********** ** *** *****.
***** *****
********* ** *******/******** ******** *** **** **************** ** *** ******** *****, ***** ***** ******* **** *** detected ** *** *** *****. ********* range *** ******* ** ****** ***** (~10-12 ***), ********** ** ********* ** travel/virtual ******** *********. ** **** ***** virtual ******** *********, ***** ********* ****** be ********* **********, ** ******* *** cross ** ** ******** ***** ** day, ******* ******.
********* *****
********* ***** ********* *********** **** ****** Guard **** * **** ****** ******* detection **** (******* ** *** *******), with ** ****** *********** ** *** testing. ********* ***** *** ******** **** ***** to ***** ****** **** ***** ******** to ******, *** ** *** ********* dwell ****, ****** ***** ******** **** regular ****** *** ** ******* ***** and *******.
Motion ***** **. ****
******** ** ****, ****** ***** ******** from ***** ***** ****** ** ***** animals *** ** ***** *****'* *********** calibration. **** ***** ** ******** ** reject ***** ****** ** ********** ******* object ****, *** ** *** ******* of ********* *****, ***** ****** ***** does *** *******.
******* ** ****, *** **** ***** looking *** *** **** ********** ****** analytic, ****** ***** ** ***********, ********** on * ****** ******* ***** *** install ** *** ****.
*************/************
**** ***** ***** ** ********** **** all **** ******* ******* ******** *.* or ******, ***** ****** **** ****** (1-2 ***** ***) *, *, *** Q ********.
*** ************ *** **** *** * series *******, *** ~$** *** *** others, *** ***********, *.*., ******* **** Motion *** ***** ***** ** * single ****** ***** **** $***.
Configuration *** ***********
*** ***** ***** ***** ************ ***** some ****** ************* *******, **** ** small ******, ******* ******, *** *****-***** object *******, ****** ***** **** ******** depending ** *** ******** (********* **** not ******* ***** *****, *** *******, as ***** ***** ******* ***** *** loiter ** ******).
*** **** ******* ************* ****** **** VMD4 ** *** ********* ** *********** in ***** ****** *******. ***** ***** two ** **** *****/**** ** ******* of ***** ****** (********* ******) ** the ***** ** ****, *** ***** the ******'* ******, ***** ****:
***** *** **** ******** ****** ***** object ******** ********, ** ** ******** versions *** ****/*. ** ****** **** process *** *** ********* ** **** analytic ** **** *****:
Beware: ******* ****** ******* ********* *****
********* ******** *** ********* ** **** 80% **** ******* ****** ******** **** increased **** ******* (*%) ** *** (20%) ** ****** ******* *****. ******** decreased ** ~**' **** ***' ** the ***** (***** *****) *** ~**' from ~**' ** *** *****-*** ** II.
Common ******: *****/*******/****
********** *******, ******* ***** *** * common ***** ***** *** ***** ***** applications, ******* ******** ***** ******. **** when ********* ***** *****, ***** ****** be ******* ** ***** ***** ******* brush *** ********* *****, ** ******** boxes *** ****** ****** ******* **** expected, ******* ***** ******.
********* ***** **** ******** **** ***** ****** caused ** ******* ***** ** ***** "Time ** ****" ******** ** **-** seconds, **** *******/***** ********** **** ****** to *****.
******* **** **** ********** ****** ******, often ********** ****** ***** *** ***** Guard, ***** *****.
*******, *** ***** ************ *** ****** **** rainfall ******* ***** ******, ******* ** **** other ****** ********* ** **** ******.
Motion ***** ***********
********, ****** ***** *** *** ***** on ***** *******, * ****** ***** during **** *******, ** **** ** other *********.
*******, ****** ***** *** *** ***** on ****** ***** ***** ******* **** as ******* ***** ******* ** *** field ** ****.
****** ***** ************ ******** ******** ** ~**-** PPF **** *** *** *****. **** results ** *** ********* ******* ******/*** widths ** ******* ******:
- *****-*** (*** **:*): ~***' **** *****/~***' ********
- *****-*** **** (*** **:*) ~***' **** width/~155' ********
Improved ***** ***** ***********
****** *******, ***** ***** ************ ******* on ******** ******** *** **** ** ~90'/10PPF. ***** ***** **** ******** ******** which ********** ** *** ******* *****, ********* ******** ******* ****** *** FOV, **** ****** *** ***:
** **** ** ** *****, ****** the ******* *** ********* ** **** ********.
*******, ** *** ** ****** ********* people ******* **** ****** ** ********, shown *****:
Solid ********* ***** ***********
********* ***** *** *** **** ** detect *** ********* ********, *** ** night, ** *** *****. ******* *** highlighted ********* ** ****** *****, *** with *** ******** ** * *****/*** which ***** ** *** ****** *******, alarming **** ****:
**** **** **** ********* ***** **** not ***** *** * ******* ****** of ******* *******, *** ******** **** alarming ** * ***** ** ***** or ****. *** "*****" ******* ** Loitering ***** ****** ******* ** ******* loitering ** ********** ***** ** * single ****** ************, ** ** *** objects ** *** ***** (*.*., ** long ** *** ******* ******* *** 120 *******, *** ******* ** **********), effectively **** ** * ******* ****** than *********** **********.
Versions ****
*** ********* ******** *** ******** ******** were **** ****** *******:
- **** *****-***: *.**.*
- **** *****-*** ****: *.**.*.*
- ***** *****: *.*.*
Comments (15)
Ethan Ace
Jun 25, 2018SB
Stian Bjerkenås
Jun 25, 2018When you get to perimeter defender, could you also include an axis thermal camera?
That is usualy what's recomended for such application, and it would be interesting to see how they perform compared to eo cameras. (given that you mount them side by side during your test)
UM
Undisclosed Manufacturer #4
Jun 26, 2018It would be really nice to see comparison side by side Avigilon, Axis and Bosch analytics. I mean same scene same situations and compare the results from 1 week testing.
U
Undisclosed #1
Jun 25, 2018Finally, we saw no issues detecting people walking very slowly or crawling, shown below
If it cannot eliminate the false alarms from other motion, this really is not that impressive.
UM
Undisclosed Manufacturer #2
Jun 25, 2018U
Undisclosed #1
Jun 25, 2018"Analytics", to me, infers the ability for the system to make intelligent decisions with repeatable or predictable results. Intelligent decisions would mean ignoring general motion and only alerting on specific objects of interest. Based on the test results here, the Axis system failed to show significant intelligence, as it had several false alarms. There is nothing to indicate it detected a "person" crawling, just that it saw the crawling as the same kind of random motion it picked up elsewhere and was unable to classify or filter out.
Such a system is, to me, nowhere near "far better than not using analytics", as it just generates random alarms and can't be trusted.
Daniel Cyr
Jun 28, 2018I think that it all depends on your application and reason of using analytics rule. If you want to go through all activity the next day to see what went on, then this analytics could help you accomplish this faster. It would at least reduce drastically the number of events compared to if you would have simply relied on motion detection.
I think that as one of the leading manufacturer of IP cameras, Axis needs to be more competitive in this aspect of their products. As per many who I have talked to that attended the Axis Summit in Jamaica, they mention that they are focusing and investing on improving their analytics. They also mentioned that most partners where not even using analytics yet so I guess some of us are just a little ahead in that sense.
Let's wait and see what this gives in the next couple of months!
Ethan Ace
Jun 25, 2018To be clear, we are not saying that it's impressive that it detects people running or crawling. It should detect those people.
It was specifically called out only because in our original test of Fence Guard, it failed to detect in these cases. I clarified that in the text.
UM
Undisclosed Manufacturer #2
Jun 25, 2018Analytics are not and will never (for a long time, anyway) be 100%. Users should expect false alarms regardless of the vendor. What noise would you receive without analytics? This is my point. If this does not make sense, please explain.
Where does it say that they analytics failed to classify the crawling person vs. 'random motion' as you say? The crawling person displayed a green box that turned red upon crossing the line. This suggests the person was detected and triggered the line. Again, not sure what I'm misunderstanding but I'd genuinely like to know.
U
Undisclosed #1
Jun 25, 2018The shadows crossing the line showed the exact same response:
Given that it failed to properly suppress the shadows and false motion, there is no reason to believe the alarm on the crawling person was anything more than accidental. Going on the results of this test as shown here, the system did not exhibit signs of being able to actually be used to detect crawling people in a manner that would not generate excessive false alarms.
Daniel Cyr
Jun 28, 2018The thing is that if you have used solid analytics like Avigilon, although they are not 100% they are VERY high up there. Avigilon has AI which over time increases efficiency or the analytic rules and eliminates some detections that at the time of install would have generated a false alarm.
At our monitoring central we monitor several hundreds of Avigilon cameras and sometimes we can go many minutes without one single event. A lot has to do with the the way these rules are configured by our integrators and the understanding of how they work.
One thing that I find amazing is that on all analytic solutions that I have tested or used, Avigilon is way ahead as the best solution but what amazes me is that it is also by far the simplest to configure.
U
Undisclosed #3
Jun 26, 2018Excellent performance! Feet planted, both hands thrust into front pockets, sporadic movement, furtive glances throughout; capturing the very essence of loitering.
UM
Undisclosed Manufacturer #5
Jun 13, 2021What is the difference between the Axis Guard Suite and VMD4?
Ethan Ace
Jun 14, 2021The main difference in configuration is that Motion Guard includes perspective setup which VMD4 does not, which improves detection somewhat. We found that it was a bit better at rejecting some false alarms, especially animals.
Fence guard is similar, but allows only tripwires/lines, not intrusion zones, and loitering guard is similar to Motion Guard except with a longer configurable dwell time.
My general recommendation is that you use Guard Suite instead of VMD4 if it's an option.
UM
Undisclosed Manufacturer #5
Jun 14, 2021