Camera Analytics Shootout 2020 - Avigilon, Axis, Bosch, Dahua, Hanwha, Hikvision, Uniview, Vivotek

By IPVM Team, Published Jan 22, 2020, 09:38am EST (Research)

[******:****** ********* ******** **** - ***, Avigilon, ****, *****, *****, ******, *********, Lilin, ******, *******, *******, *******, ******* and *******]

********* *** *** *****, ****** ** * **** ** AI-powered *******, *** ***** ********* ****** work? *** *** ** ***** *** smart ******* ******* ** *** ***-** analytics?

IPVM Image

** **** ******, ** ******* *** contrast ** ********* ****** ********* **** 8 ************* **** *** *****:

******, ** *****:

  • *** ****, ******* *** ***** ********** overall
  • *** *** * **** ********** ******** in ******
  • *** ********** ** *** ******* *** worst **********
  • *** *** *********** ******** ******* ***** alerts (*.*., *******, ***** *******, *******/******, rain)
  • *** *** *********** ******* ****** ***** alerts (*.*., *******, ******, *** *******)
  • *** **** ********* ******** ******** ******* and ********
  • *** ********** ******* **** ********** ** vehicle, *****, ***.
  • *** ********* *** ******** ***** (******* pixel ******* / *** ************)
  • **** ******** ********* ****** **** ******** had
  • *** *** ****** ** ***** ** configuration *** ***********
  • *** ***** ********* (** ***) ***** cameras ******** ******

Performance ********

***** ** *** *****, ******** *********** falls **** **** ******:

IPVM Image

  • *** **********:********'* ** ****-******** *********, ***** ***, and *****'* *********+ ******* ********* **** in *** *******, ************ ********* ***** alerts **** ****** ******* ******* *** out, **** **** *****/***** **** ******* PPF ************.
  • ***** *******:********'* ** *********, **** ********* ********, Hikvision ********** *** ********, ******* **** Learning ******, *** ******* *** *** performed ***** ******* ** *** *******, rejecting **** ***** ******, ****** **** with *** ***********. ********'* ** ********* drastically ******** ********* ******** ** *****, not ** ***** ** ** *******. Axis ********* ******** ************ ******** **** false ****** ****** ** ********** *** occasionally ********* ** ****** ******* **** as ****** ***** *******/*******. ********* ********** and ******** **** ******** **** ***** alerts ** ***** *** ***** *******. Uniview **** ******** ****** ******** **** false ****** ** ***** *******. *******, Vivotek *** ********* ****, *** **** detected ******, **** ** ****** *** vehicles, **** ****** *** ************ **** others.
  • ******* ***********:****' ****** ***** *** ****, *****'* Essential ***** ********* (***), *** ******'* Wisenet * ********* *** ********* *********, with ******** ***** ***** ******, *** a **** ***** ********* ********* **** Dahua, *********, *** *******.
  • ***** ***********:*****, *********, *** ******* ********* ************ performed ***** ** *** **********, **** increased ***** ***** *****, ******* ********* ranges, *** ***** *** ************. ***** analytics *** ****** ** ** **** in *** ** *** ******* **** these *************' ****, ** **** **** outperform ******* "****" ****** ***, *** for ******* *** *** *********** **** third *******, **** *** *** ********* options.

Top ********* ***************

******* ** ********'* **, ***** ***, and ***** *********+ ** **** ******, there *** ******* *** *********** ******* the *** ********** ***** ****** *********, detailed *****.

IPVM Image

  • ***********, ********* ******** *** *****:********'* ********* *** *********** "**** ********", requiring ***** ****** ** **** ****** a **** *** **** * ********* zone ** ****/**** *********. *********+ **** requires ***** ** ****** *** **** and ****** ****** ***/** ******* *********. By ********, ***** *** ******** **** detailed *****, **** ***** ******** ** measure *** ****** ** ******* ** the *****, *** ** ****, ******* objects, ******** ********* *** ****** ** the ****** *****, *** ********, ***** length, *** ******** ****** ***** (********* in *** ***** *****).

  • *** ***********, ********* *****:***** *** ********** **** ******** *****, including **** ****** *** ******** *** metadata. ********'* ********* ********* **** **** Avigilon ******* ******, **** ** ************* with *****, *******, *********, ** ******. Dahua *********+ **** ********* *** ******* Optix' ** *******, *** *** ********, Exacq, ** *******.
  • ****, ********* *****:***** ****** *********+ ******* ** **** lower ****** **** ******** *** *****, with **** ******* **** ****** ******* for ~$*** ******, ******** ** $***-*** at * *******.
  • ********* *********/*** ************ *******:** **** *****, ********'* ** ******* dropped ********* ******** ** *****, *** their ** ****** ** *** ****** from **** **********. *** ***** *** analytics **** ****** ******* *** ************, from ~*-** ***.

******* ** ***** ******, ***** ********* provided ******* ***** ***** ***********, ********* most ****** ****** *** ******* *******, such ** ******* *****, ***** *******, rain, *******, ******, ***.

Deep ******** ********* ****** ** *******

** *** *******, *** **** ********/**-***** analytics ********* ***** ******* ** ****, dominating *** *** *** **********, **** Avigilon's *** *** ***** *** *** only ********* ***** **** *** **-***** in ***** *** ********. ** ****** that ** ** ******* ************ ********* to ******** *** ********** ********, ** will *** **** ***** ********, **** even *** ****/"*****" ** ******** *********** superior ** **** ***-** *********.

Outdoor ***** ***** *********** **********

**** ********* ****** *** ****** **** multiple ***** ***** *******, **** ** foliage/brush, ***** *******, *******, *** ****, detailed *****, ****** ******** (**** **/**), Bosch ***, *** *****'* *********+ *** not ****** **** ****** ** *** of ***** *******.

IPVM Image

******* *** ********** *** **** ****** issue ** ******* ***** ******, *** in **** **** ********* **** *** triggered ** **. *******, *********'* ********* Analytics *** ******* ********* *** ******** from **** ******** ****** **** ******* was ******.

IPVM Image

**** ** *** *** **** ****** source ** ***** ****** ** ********* outdoors, ******* ***** ****** ** **** (Motion ***** *** ****), ***** ***, and *****, ******, *********, *** ******* intrusion *********. **** **** ***** ****** were ****** ****** ** ******** ******** down *** *******' *****, *** **** alerted ** ***** ******** ******.

IPVM Image

****, ***** *** ***** ******* ************ triggered ****** ** **** ****, ***** EVA, ***** *** ******* ********* *** Deep ********, *** *** ********* *********, including ********** *** ********.

IPVM Image

**** ****** ***** *** ********* ** headlights ******* ** ******* ** *** scene (**** ** ****** *** *******), e.g., *** ******* ***** ** *** example *****.

IPVM Image

** ******** ** ***** ****** *******, Uniview's ********* ********* **** *** **** ones ** ***** ** ******* ****** by ****** ******* ********, ***** *****. No ***** ****** *** ***** ******.

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

** **** *******, **** ********* ** not ***** ***** ** ****** ****** sources, **** ** ****** ***** ****** on *** ***, *********** *** ********, and ******* **** ** ***** *******, with **** ******* **********.

IPVM Image

******'* ******* * ********* **** ***** with ***** ******, ******** **** ****** were ****** ** ** ***, **** as **** *****:

*** ******** **** ****** **** ****** on/off, ******* ***** **** *** ****, below:

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

****, *** *** *** ********* **** able ** ******** ****** ******** ******* and ********, ******** *****. ************, ************** issues (** ***** ***** ********** ******* as ******* **. *****) **** ****, but ******* ** ****. *******, **** analytics ******** ****** ****, ********* ** constant ******.

IPVM Image

*** **** ******** ***** ****** ******* subjects ** *** ***** *** *********'* AcuSense. **** ******** *** *** *****, indoors *** ***, ********** ** ******* the ******* *** ****** *** ***** of **** ** ******.

**** ********* ******** *** ****** ** detect ******** ******** ****** * ******** setting *** *******.

IPVM Image

*******, ***** **** ******* ******* ******** subjects ** ** ********, ** **** resulted ** ***** ****** ** ******* in *** *****, ***** ****:

[optimize output image]

** ********* ********, ****** ******* ******* the *****, ****** ** *** ********** and ******* ****** *** ****** **** frequently ********** ** ********. **** ***** lead ** ********** ****** ********** ** users ****** ** ****** **** ******.

IPVM Image

*******, ** ******* ***** ** ***** analytics (******** ***, ***** *********+, *** Hikvision ********), ****** ******** ** *** scene ********** ********* ******. ***** ****** were *** ******* ** *** ***-** analytics, ** *** ******* **** ****** not **** ** ******.

IPVM Image

PPF ************/********* *****

*** ****** *** ******** ** ********* tested *** ~* *** ** **** Perimeter ********, *** *** *****. *******, Perimeter ******** *** ***** ** ***** alerts **** ******* ******* **** ** brush *** ***** *******, ** ****, unlike ******** *** ***** ***, ***** required ~* *** (****** *** ******** in **** ******, ********* *****) *** did *** ****** **** ***** ******.

*****'* ***/*** **** **** ******, ** 8/9 ***, ******** ** ********'* ***, Axis ****** *****, ********* ******** *** DeepinView, *** ******* ***, *** ********* 10-12 ***.

********'* ** ********* *********** * *********** drop ** ********* ***** ***** * lux, *** ***** ** ***** *********. PPF ******** ***** **** ~* ****** the *** ** *** * *** 8 ** ****** ** **+, ******* detection ******** ***************. *** *** ****** performed ********** ******, *** ***** *********** nearly * **% ******** ** ********.

IPVM Image

Avigilon **

******* *** ******* ***** ******* *******, with ************ ***** ***** *** *******, Avigilon ******* **** *** **** ** detect *** ***** ** *** **** nearer *** ****** **** ***** ******. For *******, ** ~**' *****, *** subject ** ******* *****, *** ** alarm ** *********:

IPVM Image

** ********, ********'* *** ******* ******** at ~*-** *** **** *** *** night:

IPVM Image

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

*** *********** ****** ******, **** **** supporting ******** ******, ****** ********** ******** boxes, *** ****** **** ********** *** other ***, ********** ** *** ***** below:

  • ******* **** ******** ******:*******'* **** ******** ****** ********* **** not ********* ** *** ***.
  • ******** ******* ******:********'* ****-******** ********* **** ********* **** in ***** *** ******* ****** ***, Genetec ******** ******, *** *********.
  • ***** ***/**** ********* ********:***** *** *** **** ********* ******** were *** **** ********** ******** ********* tested, **** ******** ***** ********* ** multiple *****
  • **** ***** *****/**** *** ****** ****** supported:****' ***** ********* *** ******'* ******* X ********** ****** ** *** ***** tested ****** *** ******** ******* ****** (but ** ******** *****).

IPVM Image

Axis **/***** *** *** ***********

****' ********* ******** *** ***** *** were *** **** ****** ********* ***** integrated ******** ***** ** **** ** events ** ***** ***** *****, ****** the ***** *********** ****** ********. ********* Defender ******** *** **** ******** *****, but *****, ** **** ** * green/red ***** ********* ** *** *** right ** *** *** ******, ***** here ** *********:

IPVM Image

** ********, ***** *** ******** **** bounding ***** ****** ******** *******, **** no **** ***********, ***** **** ** Genetec ******** ******.

IPVM Image

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

**** *** ** *** ****** ********* have ******** *******/******* *****: **** ***** Suite (~$**/******* *** ***-* ****** **** models) *** **** ********* ******** (~$***/*******). Others *** ******** ** ** ********** cost ** * ***** *************' *******.

*******, ****** ***** *** ** "** additional ****", **** ************* **** ******* analytics ** ****** ****** ******, **** as ********'* **/** ****** (*** ******* mini **** *** ****) ** ******'* Wisenet * ******* * (***** ******** VMD ****). ** ********, *****, *********, and ******* ******* ********* ********* ** nearly *** ******, **** *** **********, with *** ******* ***** **** ***********.

Comments (48)

You have 'Axis Perimeter Defender' listed differently in the rankings table (Average) versus the summary text (Above average). Also, will you be adding a list of tested models?

Agree
Disagree
Informative
Unhelpful
Funny

Thanks for pointing that out, I moved perimeter defender up to correctly show its ranking in the chart. If you would like to see the models tested they are in the individual reports linked at the top.

Agree
Disagree
Informative
Unhelpful
Funny

Ah yes, paying closer attention I see now that this is a summary of the previous tests, as opposed to including new testing. What do you know about the 2nd generation Accusense models, if anything? I see that both Hikvision Global and Europe show several more Accusense models than what is available in the U.S., including 4K versions such as this one:

DS-2CD2686G2-IZS

Other than the "G2" in the model name (which may mean nothing since there are various "G0" models), I see no indication of whether it is 2nd gen. Hikvision is claiming better false alert rejection for the 2nd gen, and in fact they also have a firmware version note for the existing Accusense models claiming that it reduces false alerts.

Agree
Disagree
Informative
Unhelpful
Funny

We're using more than >2000 pieces of H4 for construction site security and are generally very happy with the results of the VCA. Now that H4 is not available anymore we switched to H5 and compared both cameras (in different megapixel versions) right next to each other. The unexpected outcome of the test was, that the H5 did perform significantly worse than the H4 during the night (with external IR lights) in detecting people in different clothings walking, running or crawling through the scenery in different distances (15,30,45,60.75 meter).

When discussing the outcome with our supplier we got the following statement back:

"Here are our conclusions and let's start with the most important finding, which is the right perception of the H5 cameras' confidence rate. For a good reason, this value is critical to find out where the technical boundaries of our system lies. After evaluation of the clips, we quickly became aware that we missed to properly communicate the right background information on how to actually read this value correctly. In first glance, it looks as H5 confidence rates have no improvement to H4, but H5 is actually using a completely different scale. Simply put, a low H5 confidence rate equals to a mediocre H4 value. To put it in clear figures, a 50% H4 confidence rate equals to 10% in H5.

Reviewing the footage again with this information, we can see that H5 gives us significantly better object detection results and thus, equally important, way less false alarms."

Before running the next extensive test series, I'd like to understand if someone of you faced the same issue. Furthermore I'd like to understand what do you think of the statement of our supplier.

Thanks a lot.

Agree
Disagree
Informative
Unhelpful
Funny

In first glance, it looks as H5 confidence rates have no improvement to H4, but H5 is actually using a completely different scale. Simply put, a low H5 confidence rate equals to a mediocre H4 value. To put it in clear figures, a 50% H4 confidence rate equals to 10% in H5.

sounds like bunk to me. what is the "different scale" exactly?

when a h4 camera says 50%, what's the reality? is it right half the time or one out of ten?

Agree
Disagree
Informative
Unhelpful
Funny

ds like bunk to me. what is the "different scale" exactly? when a h4 camera s

They said that 50% of Confidence level of H4 is the same level of confidence of as 10% shown in H5.

Agree
Disagree
Informative
Unhelpful
Funny

We've contacted Avigilon for clarity on this. I'll post back here when they get us information.

Our test results our pretty different from yours. H5 was far beyond H4 for nighttime detection. In our H4 tests, subjects could walk back and forth, clearly visible, and not be picked up at all.

Agree
Disagree
Informative: 3
Unhelpful
Funny

Did you get a feedback from Avigilon on the scale, yet?

Agree
Disagree
Informative
Unhelpful
Funny

Avigilon has replied, the confidence ratings of the H4 and H5 arent directly comparable and the only accurate way to compare the two is to measure false positives and missed detections:

While "confidence" conceptually means the same thing (how certain is the classifier that a classification is correct?) for both H4 and H5, they are not directly comparable. H5 leverages a more powerful convolutional neural network based classifier that was trained with a larger data set than the H4 classifier. Confidence values are calibrated such that at the lowest confidence, we maximize detection rates while keeping false positives at an acceptable rate while at the highest confidence, we minimize the false positive rates while keeping detections at an acceptable rate. A more powerful classifier (H5) allows more correct detections across all confidence values compared to a lesser classifier (H4). Thus, because of the difference in the operating range of the classifier, it is difficult to directly compare an H4 classification confidence with that of the H5.

Agree
Disagree
Informative
Unhelpful
Funny

All right. Thank you.

My quote originally came from Avigilon as well, when they tried to explain the poor detction of the H5 compared to the H4. They have been part of the test-setup and couldn't explain the outcome, yet.

However, if I understood your quote right, we can compare the H5 and H4 by setting the confidence levels to a certain rate (say: 30%) and count the missed detections for both.

In exactly this test, the H4 performed by far better than the H5. And this is actually my problem, because H4 is no longer available and H5 missed an awful lot of people.

Any thought what we can do ?

Agree
Disagree
Informative
Unhelpful
Funny

Has there been any thought given for testing the various analytics platforms with thermal cameras?

For crawlers, was any testing done with prone/flat targets rather than on their hands and knees?

Agree: 1
Disagree
Informative
Unhelpful
Funny

Thanks for that comparison, we are working with Uniview and recently tested their Alphaview solution with much (much) better results, might be pretty useful for us to see that added to comparison...

Agree
Disagree
Informative
Unhelpful
Funny

for that comparison, we are working with Uniview and recently tested their Alphaview solution with much (much) better results, might be pretty useful for us to see that added to c

Which Alpha View did you text, yet ?

Agree
Disagree
Informative
Unhelpful
Funny

i have a IPC6858SR-X38UP-VC which is part of alphaview and its really disappointing. running it for the last 2 months now and after all the issues i highlighted to uniview. they tell me the NVR which can use the potential of the camera is not available yet.

just to clarify, the issues i am highlighting is relating to vehicle and object identification , poor or non existent autotracking once detected and others

right now, i am looking at the camera locked on to a stationary white vehicle for the last 30 min and not returning to home position.

Agree
Disagree
Informative
Unhelpful
Funny

Hi Gents - on the VMS integrations list the colour (correct spelling) of the + is blue in the table but orange in the key - or that's my assumption.

Minor detail but jarring to the eye !!!

Great report thank you.

Agree: 1
Disagree
Informative
Unhelpful
Funny

We had to pull some Hanwha for outdoor coverage, way too many false alarms and hugely unreliable. Tech support wasn’t Mitch help and we just moved the client to Avigilon with successful results.

Agree
Disagree
Informative: 2
Unhelpful
Funny

Hello,

Have you done any kind of test for "group of people"?

Any time a group of 4-5 people, that are walking together, is being detected as a vehicle or a people?

Agree
Disagree
Informative
Unhelpful
Funny

Some interesting context on the state of analytics. I have three Hik "Performance" series and two "Smart" series (late 2018 release?), and one Eufy video doorbell. (No monthly subscription cost.) Thus far the $160 retail list but selling for less than $130 Eufy doorbell camera is blowing away the "Smart" series Hiks on face and human detection. The three Performance series cams may as well not even bother with a "Face Detection" option, since it does not work at all even at maximum sensitivity and with a face 5 or 6 feet in front of the camera. Meanwhile the 8 MP smart cam will miss faces consistently, and then the 4 MP smart cam will occasionally trigger on a scene with no motion at all with IR on.

I can't locate any teardowns of the Eufy so I don't know what chips are used, but it would be interesting to test it in short range scenarios against the much higher dollar cameras. It's identifying moving people 15 feet or more from the camera, and I have yet to be able to fool it except with a human shaped boxing dummy. Lifelike enough you would want to get a notice if it walked up to your door somehow.

Agree
Disagree
Informative
Unhelpful
Funny

In taking another look for the "unicorn" of 4 MP or more and good analytics at $700 or less, it appears like it's time for an updated test of the Hanwha Wisenet X models. Their firmware has been updated several times since your previous test, and they claim improved detection and other changes. And as of now, B&H has the 5 MP XNV-8020R for $475, which appears to be simply a fixed focal version of the XNV-8080R included in your test.

https://www.bhphotovideo.com/c/product/1336793-REG/hanwha_techwin_xnv_8020r_5mp_ir_outdoor_vandal.html

In addition, it appears as though here in your summary overview, you only used the weaker performing "Intrusion" mode on the X series for comparison against the other brands. Correct me if I'm wrong, but from your specific Hanwha test report linked above:

"While Hanwha's intrusion analytics and VMD worked poorly, with numerous false alert issues from many common sources, switching to loitering with a short 1-2 second delay worked very well as a workaround / alternative, eliminating many of these false alerts."

Heavy rain on the dome was a problem you noted even for loitering, but this is not a problem for covered locations. That leaves only reflections and some light changes as false alert problems. At $475 for a non Hikua option, those false alerts may be an acceptable trade off even if they have not improved with newer firmware.

In reviewing the firmware release notes, version 1.20 added a time delay to intrusion of up to 5 seconds. This alone should warrant a retest, and version 1.40 claims "Improved motion detection algorithm to increase the detection rate of recognized objects".

The new 1.40 firmware was on the XNV-6081Z in your recent test of that model, but that test did not look at analytics. (!) I'll suggest again that's it's time for every test of every model to look at all aspects of performance. Of course that takes more time, but that's how things are done in most of the product testing world, and it will make things quicker and easier for your readers to find the information they seek.

Agree
Disagree
Informative
Unhelpful
Funny

We tested with updated firmware before republishing this. Performance didn't change. The only thing that changed was the inclusion of the new minimum duration setting in Hanwha models.

In addition to this test, we also have one running in our office covering a hallway, and I can tell you that Intrusion triggers every time the lights are turned on or off (they're on an occupancy switch).

Hanwha's AI models are expected in Q2 and we plan to test them as soon as they're available.

Agree
Disagree
Informative: 2
Unhelpful
Funny

Thanks for the feedback. So you didn't/don't see an improvement using 2 seconds (or 3) on Intrusion? And using only Loitering with a 2 second delay as in your report, and turning off Intrusion, the X series does not make it to above average?

The wording of your initial report specific to the X series, to me, indicates that it's not up there with the best models, but it is above average when using Loitering only. It does seem odd that adding the delay to Intrusion made no difference there.

Agree
Disagree
Informative
Unhelpful
Funny

Chris, this report more accurately reflects the performance of the Wisenet X analytics, which were average even with the dwell time on intrusion/loitering. We tested the newer firmware in the scenes at our new facility and they would false alarm on swaying branches of an evergreen or lights being turned on and off in a hallway.

Agree
Disagree
Informative
Unhelpful
Funny

That does not really answer my most recent question on your initial 2018 Loitering feature results versus Intrusion with the newer firmware. Loitering is alerting on swaying branches/shadows? Loitering alerts the same as Intrusion with sunlight to cloud changes? If yes on these, is it fair to say the initial report had different results or criteria such that the performance was overstated?

The direct quote from the 2018 report contradicts what you just replied with:

"Using Loitering with optimized settings (2 second delay, ~70 sensitivity), indoor and outdoor sources such as small animals, foliage, shadows, and light changes were effectively eliminated."

(Can't get the block quoting option to work only on selected text for some reason...)

The just posted favorite video analytics article bolsters my impression that the X series is above average as shown in your initial report, whether based on settings various integrators are using or whatever. Why would Hanwha come in 2nd if it's best analytics models are only average? Not trying to be difficult, but the complete view of information here is not really clear.

Based on my initial comment above and your reply late last month, my impression was that new testing was not done. I understand now you have done new testing with newer firmware, but you should make that very clear in this newer report, as opposed to linking to the old one and then reinforcing the idea that no new testing was done via your reply above.

Agree
Disagree
Informative
Unhelpful
Funny

Why would Hanwha come in 2nd if it's best analytics models are only average?

Like how Axis has done well in the past (e.g., see Favorite Video Analytic Manufacturers 2018), it's a function of people making use out of what cameras they already use. For example, Axis analytics have never been comparatively strong but lots of people use Axis, so a certain percentage will make do with their analytics.

I'll level the specific technical discussion here to you, Ethan and Rob.

Agree
Disagree
Informative
Unhelpful
Funny

Can you tell me when will the test be conducted?

Agree
Disagree
Informative
Unhelpful
Funny

We have two test reports from Hanwha's AI models:

  1. Hanwha AI Analytics Camera Tested, which covers basic detection, false alerts, range, etc.
  2. Hanwha AI Object Detection Tested, which covers object metadata accuracy and search

They worked pretty well, resistant to most false alerts and with (mostly) accurate metadata. They did occasionally misclassify objects as vehicles or people and false alerts from rain were possible.

We ranked them in the successor to this 2020 shootout in Camera Analytics Shootout 2021 - Ava, Avigilon, Axis, Bosch, Dahua, Hanwha, Hikvision, Lilin, Meraki, Mobotix, Rhombus, Uniview, Verkada and Vivotek.

Agree
Disagree
Informative
Unhelpful
Funny

Since the release of Genetec Security Center 5.7 SR6 and onwards about 18 of the Avigilon Deep Learning Analytic Events from the H4 and Pro Series pass through to Genetec.

Some details and screenshots:

Dropbox - Support for Avigilon Cameras & Analytics.pdf - Simplify your life

Although this is not documented in any of the release notes it apparently works as I have done a few tests with the H4 MultiSensor.

Avigilon Unusual Motion and Appearance Search are not supported. Not sure why Unusual Motion isn't supported - maybe because it may require some additional analytical support from ACC and may not be totally edge based.

Agree: 1
Disagree
Informative: 1
Unhelpful
Funny

UMD is not used to trigger any events at this time its only used to filter events in recorded video.

Agree
Disagree
Informative
Unhelpful
Funny

Thanks for the heads up on this. I noticed recently they added this point in the latest firmware release notes for H4 Multisensor:

So theoretically, third party VMSes should be able to draw bounding boxes. In practice, I haven't seen anyone doing it.

Agree
Disagree
Informative
Unhelpful
Funny

Hi, everybody
I'm new to the business. I live in Italy. I'm interested in a Dahua nvr with analytics+. Unfortunately outside the United States Dahua does not use the same names and serial numbers for nvr. Analytics+ does not exist:

https://www.dahuasecurity.com/products/All-Products/Network-Recorders

Having read the comparative test and found that analytics+ had an excellent price-performance ratio I would have liked to test this product.
Can you help me find out which is the corresponding product?
Franco

Agree
Disagree
Informative
Unhelpful
Funny

Hi Franco, and thanks for your first comment!

There's no public cross reference for international to US models, unfortunately, but looking at specs, I believe the N45DL7Z we tested is equivalent to one of the IPC-HDBW5442E models. There are several variants, ZE, ZHE, Z4E, and I'm not totally sure the differences of each, but they are all in the Pro-AI series and should include Analytics+.

Agree
Disagree
Informative
Unhelpful
Funny

Oh and one further clarification: Analytics+ itself is a US marketing term. In the rest of the world they simply refer to it as "Perimeter Protection".

Agree
Disagree
Informative
Unhelpful
Funny

Thank you very much Ethan!

franco

Agree
Disagree
Informative
Unhelpful
Funny

A last question: if I buy an analytics+ camera do I need an analytics+ nvr to let them work properly?

and on the opposite side, if I use an analytics+ nvr do I need analytics+ cameras?

thanks in advance

Agree
Disagree
Informative
Unhelpful
Funny

The camera and NVR can both work independently of eachother, you can use one or the other for analytics+.

Agree
Disagree
Informative
Unhelpful
Funny

Thanks Rob

Agree
Disagree
Informative
Unhelpful
Funny

I've tested Dahua Analytics+ (equivalent in EU), instead of Avigilon H5A. It's a real bullshit : trigger everything with rain and wind (even with line, humain/vehicule, etc). I was ready to payback customer, so I've changed to Avigilon : I can now sleep peacefully.

Agree
Disagree
Informative
Unhelpful
Funny

Update:

Added Milestone as supporting Avigilon event integration and bounding boxes.

Agree
Disagree
Informative
Unhelpful
Funny

Rob, appears that the text was updated but not the table.

Agree
Disagree
Informative
Unhelpful
Funny

Kambiz, thanks for pointing that out, I updated the graphic to show Milestone support for Avigilon event integration and bounding boxes.

Agree
Disagree
Informative
Unhelpful
Funny

What is your opinion on IR on board with analytics to detect persons in an area? We now have a lot of projects where we need to choose for a camera with a good on board analytics. We do have some experience with Bosch IVA but also want to test the Avigilon H5 series. But what about spiders and their web? Can we still assume that it will be a problem in the Avigilon series and that it will not work during night because of the IR reflection? Thanks a lot.

Agree
Disagree
Informative: 1
Unhelpful
Funny

Update:

Added Uniview Deep Learning Camera to the shootout ranked as above average. While Uniview did not miss people walking or running and had no false alerts on common sources such as outdoor swaying foliage, shadows, rain, or indoor major light changes, it did frequently false alert on large animals in the scene.

Agree
Disagree
Informative: 1
Unhelpful
Funny

Great Article but I keep watching what people think are quality analytics by the "BEST CAMERA COMPANIES and I did not see one thing in this report that we could not do more than 10 years ago.

We even hid under black tarps at US Military Show in VA in slow crawl at less than 4 inches per 10 minutes and analytics worked successfully. Even did same at major facilities in Israel and DC.

We could do traffic in downtown Tel Aviv and capture bikes, cars, tell if they were cabs or police cars. Colors, numbers, names all speeding on the highway. Our analytics, which we designed ourselves were in cameras, dual processors, could do multiple analytics at one time very successfully. Our video we showed to our clients were live and real, not from a lab.

Worked in heavy show at airports and captured the people and vehicles going to and from, setting right and wrong direction and analytics. Perimeter detection at the most

In hurricane weather in Florida, only 2 false alarms incorrect at largest electrical company in FL, with all kinds of debris flying.

Perimeter detection was better than what I see in this report.

Tell male or female- 90% accuracy, people in uniform or without.

AS an FYI, none of the above manufacturers is the company that had these analytics and was not included in the analytics that Avigilon swooped up when trying to buy the best analytic companies, I think their first was Object Video and was the leading joke in the industry with more false alarms than positive alarms. I really think they were the major reason people did not trust analytics for a decade. Avigilon bought them and their patents, seriously. Not one time did I see them beat us in a test, also not EVEN Agent VI. A more hybrid solution and still around today.

Analytics are great if you have someone watching the screen or issue happens and emails , texts with alert with video.

The 3rd stage of analytics is just now starting and then we will see great analytics. The analytics today are really not the best in LEARNING ANALYTICS yet, but maybe soon will be.

Agree
Disagree
Informative
Unhelpful
Funny

Do you also have a statistics on how much the probability is for a false alarm in specific test cases?

Or how many false alarms per typical day, week or month are triggered per brand? Are there some statistics like that available?

Agree
Disagree
Informative
Unhelpful
Funny

It is possible to capture footage from a camera and run said footage through another camera's out of the box VCAs, but it is not the simplest of tasks. There are also a lot of setting tweaks and third party software packages that make the comparison even more difficult.

In todays environment it is possible to slap a HikVision Camera on the wall and run third party analytics on the triggered email images/RTSP Stream and get similar performance to an Avigilon System (Not that I would recommend this).

Agree
Disagree
Informative
Unhelpful
Funny

Or how many false alarms per typical day, week or month are triggered per brand?

Tolga, thanks for your first comment! Good question.

On purpose, we do not do 'per typical day' metrics. The reason is that most analytics fail under certain types of phenomena, i.e., small animals or rain or headlights. Ergo we make clear what phenomenon impacts each analytics. Maybe you are in a desert or a rain forest or have lots of dogs or no dogs. This will impact how much a brand's analytics deficiencies will cause practical problems.

Our advice would be to look at the issues we found with the brands you are considering and then estimate how likely those issues will be of a practical concern to you.

Agree: 1
Disagree
Informative
Unhelpful
Funny

I've tested Dahua Analytics+ (equivalent in EU), instead of Avigilon H5A. It's a real bullshit : trigger everything with rain and wind (even with line, humain/vehicule, etc).

I've changed to Avigilon : I can now sleep peacefully.

Agree
Disagree
Informative
Unhelpful
Funny

We don't have this experience. We always prefer Bosch IVA camera's for human detection but using Dahua Pro or Lite AI with seperate lighting from a IR spot give also good results and not tha much fals detections.

Agree
Disagree
Informative
Unhelpful
Funny
Subscribe to IPVM Research to read the full report.
Why do I need to subscribe?
The IPVM Research Service includes products tests and shootouts plus competitive and financial analysis, helping decision-makers better evaluate purchasing, partnering, developing, and/or competing against companies in physical security.
Already have an account?
Loading Related Reports