Avigilon H5A Analytic Cameras Tested

By Rob Kilpatrick and Derek Ward, Published Oct 07, 2019, 10:19am EDT

Avigilon has released its H5A analytic cameras, claiming to "detect more objects with greater accuracy even in crowded scenes."

IPVM Image

We tested the 1080p and 4K models of the H5A line to see how they perform compared to the H4A line of analytic cameras, tested in our report Avigilon "Self-Learning" Analytics And VMD Tested.

Inside, we answer the following questions:

  • How well do they perform against common outdoor false alert sources such as rain, animals, and foliage?
  • Are human and vehicle classification accurate?
  • What distance/PPF level is required for detection?
  • How do image details in day, night, and low light compare to H4?
  • How does WDR performance compare?
  • How well do they integrate with third party VMSes?

Note: Appearance Search Coming: This test covers fundamental analytics and image quality. We will be releasing our findings on Avigilon's Appearance Search and UMD in an upcoming report.

*******

***** ** *** *****, Avigilon's ** **** *** several *********:

  • ****** ***** ***** *********:****** ******* ** ******/******* false ****** **** ******* such ** ******* *******, small/large *******, ***** ****, drastic ***** *******, *** shadows ** *** *****.
  • **** ******** *********:** *** *****, *** H5 ******* ******** **** ~9-10 *** *** ********* of ******, *** ** the ****** ************ ** analytics ** **** ****** (*** *** ********). ****** **** ******** at **** ***' ******** in * **°+ ***** of **** ***** *** 4K ** ******.
  • ******** ***** ******* ******** To **:******** ** ******** ** models, *** ** ******* produce ****** ******* ** all ******, *** *** night.
  • **** ** ************:*** **'* ********** ** illuminator ****** *** *** scene, ********* ***** ***** on ******** ** *** scene.
  • ***** ******* ***:** *******' **** ******* performance *** ****** **** average (*** *** *** Mega ********), ******** ******** to ** ****** ***** were ******* ** ***** in *** *****.

*******, ***** **** * few *********, ** ****:

  • ***** ****** ** ****** Vehicles:****** ******** ** *** scene ********** ********* ******** alerts, **** ***** **** stopped ******, ******* ******* alerts. ******** **** **** issue ****** ** ***** in ******** ********.
  • **** ***** *** ***** Performance (~* ***):**** ** ******* ******** weak ******* ** *** light (~* ***) *** to **** ****** ** digital *****. ****** **** brighter **** *** ** cameras ****** (**** *****) though **** *** *** deliver **** ****** ******.
  • ****** **** ******* ********:** ***** ***** *.*** bitrates **** **** ******** to *********** ******, **** 4-5 **/* *** *** 4K ******. ** ********, competitors' ***** ***** ******** were ********** ***** * Mb/s.

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

********'* ** ****-******** ********* outperformed ***** ****** ********* we **** ******, **** false ***** ********* ******* to*****'* *********+ ****************'* ***** ****** *********, *** *** ********** in *** *****.

*******, ******** *** **** to ******** ********* ******* and ******** ********, ** issue *** ******* ***** analytics. ************, ** *** not *********** ******** ** animals ** ****** ** vice *****.

*** **** *********** ********* issue ** ***** *** the ********* ** ********** vehicles, ***** ******** **** will ** ******** ** firmware **** **** (** will **** *** ****** when *********).

Better ********* ** ******* ******

*** ** *******' ********* were **** ** ********** detect *** ******** ******** objects ** **** ****** with ******** ***** ********** simultaneously, ***** ** ****** analytics ****** ******* ********.

*** *******, ** *** image *****, ****** ** bikes *** ********* ********** (both ****** *** ****) along **** ******** ******** in *** ***** **** objects ***********:

Accurate-Classification-Of-People,-Vehicles,-Bikes

**** **** ***** ******** are ********* ******* **** vehicle ********* *** ****/***** use, *** *** ** specifically ******** *** *************** ******. ******** **** **** plan ** ********* ***** object ******* ** **** analytics ** ****** ******** but **** *** **** a **** **** **** might ** *********.

Main ******** ***** — ******** ****** ** ****** ********

** *** *******, ****** vehicles ** *** ***** frequently ********* ******. **** means **** ***** ******** regularly ********* ****-**** ****** using *******'* ******* ********, as *** ****** *** near ********** ******** ***** parked ** *** ******'* field ** ****. ***** alerts ***** *** ** completely ******* **** ******** changes, ****** **** ***** be *******.

IPVM Image

******** ************ **** **** is * *********** **** the ******* ******** ******* and **** ** ********* a *** ******** **** fixes ****** ** ****** vehicles.

Accurate ****** ********* – ******* *** ******** ********

****** *******, ****** **** not ****** ******* ******* or ******* ******* *** scene **** *** *** night.

IPVM Image

Partially ******** ******** *******

******** ** ********* **** detected *** ******* ******** behind ******* **** ********* obscured ***** **********, **** as *** ******* *****.

Accurate-Detection-Of-Partially-Occluded-Objects

Common ******* ** ***** ****** *******

********** **** * **** of ******* *** ****** sources ** ******/******* ***** alarms **** *******, ********* those ***** ********* **** other *********, **** ** heavy **** ** *** scene/raindrops ** ***** ** the ****, **** *****.

IPVM Image

** **** ** ***** and ***** *******, ***** here:

No-False-Alerts-On-Animals

*******, ****** ******* *** shadows ********* ** ***** alerts ** *** *****.

No-False-Alerts-On-Foliage

Faster ********* *******, *********** ******* ** *****

******** ****** **** ***** H5 ********* ****** ******* faster **** **, ********** detection ** "****** *********." However, ** *** ***** we ***** ******* ********** in ***** ******* ** and **, ** **** 100-200ms. **** *** **** a ********* ********** ** some **** *****, *** in *** ***** **** H4 *** ** ******* detected ******* *******, **** vehicles ****** ** ~**-** miles *** ****.

Nighttime ********* ******* ********

**** *** ******** *.*-***-***-** and *** *.*-***-***-** ******** people ** ~*-* ***, which ******** ** * ~210' ********* ******** ** the *.*-***-***-** *** * ~115' ********* ******** ** the *.*-***-***-** ** **** day *** *****.

**** ** * *********** improvement **** *** ******* which *********** ******* ********* distance ** *****, ********* ~25+ *** *** *********.

IPVM Image

Image ******* ************ ******** ** **

********'* ** ****** **** improved ***** ******* ******** to ******* ** ******.

*** *******, ** ******* images (~*** ***) ** near *****, *** *** 4K ****** ******** ******* details ** *** ******* than *** ********** **, which ******* **** *** somewhat ****.

IPVM Image

** ****** *****, ** the ** ** ******, the *******'* ****** ******* are ******* *******, *** his **** ******* ******* by ********* ** *** H4.

IPVM Image

Improved ****/** *********** (~*.** ***)

** ****** ********** ** performance ** * **** (~0.02 ***) ****, ***** here. *** ** *******' illumination *** **** ****** the ***** ** ****, with ************ **** ** very **** ***** (distance), ********* *****.

IPVM Image

**** *** ******* ***********, the ****** ****** ******** and ** ***** ** compensate, **** ** ************ until *** ******* ******* very ***** ***** (~*' distance).

Overexposure-At-Very-Near-Range-((5')

** ** ******* *** much ******** **** *** H4 *****, **** ***** subject ******* *** **** exposure ** *** *****. The ** ****** **** not **** *** **** detail ** *** ******* at ***, *** ** high ****** ** ******* noise. *** ***** ** camera ******** ****** ******* of *** ******* ** this *****.

IPVM Image

** ~**' ********, *** H4 ******'* ***** ** more ********** ***** ***** details *** ***** ******* in **** ** ******.

IPVM Image

*******, ** ~**', *** subject ** ****** ********** in *** ** ******, but ***** ******* ******* in **** **'*.

IPVM Image

Weak ***** *** ***** ***********

***** ****/** *********** ** solid, *** ** ****** produced **** ***** ******* in * ~* *** scene, **** ***** *** artifacts ******** *******.

IPVM Image

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

** ****** *** *********** against ** ******** *** door, **** ******* *** ~2,500 *** ******** *** ~100.

IPVM Image

** **** *****, *** H5 ******* ******* *********, with ***** ******* ** the ******* ******* *** backlit **** ***** ****** are ********** ****** ** the ********, ****** ***** recognizable. **** ******* ********** the ** ** ****** which *********** *** ********** and ***** ******* ******* of *** *******.

IPVM Image

** ******, *** ******* are ****** *** ******* as *** ******* ***** the ****, *** ******* usable ******* ** *** subject ** ** ******.

Above-Average-WDR-Performance-In-Motion

Bandwidth **********

** ******** **** ****** than ******* ******** ** other ** ****** ** have ******, **** ******* birates **** * **/* common *** ****** * Mb/s ******* ** ***** with ** **. ** contrast

***** ************ **** ***** with ********'* **** ***** codec ** ***** ******* settings.

IPVM Image

Camera/Analytics ***********

** *******' *.*** ****** was ****** **** ** Avigilon ******* ****** *** Genetec ******** ******, ****** H.264 ****** ******* ****** in *** ***** ******. In *** *****, ********'* analytics ********** **** *********, Genetec, *** ***** *** Avigilon ******* ******.

****: (**** **** "****** **" these ****** ** ***** parties *** ******* *).

IPVM Image

Installation ********

*** ** ******* ******* two ***** ************ ********. First, **** ******* * mounting **** **** ** integrated ***** ***** ****** the ****** ** ** hung ***** *********** *** made, ******* ** **********' hands.

******, *** ******* ******* a ******-******* *** **** that ******* * ******** adapter. ********** *** ******* to **** ******** ******* to *** *** ***** the ****** ***** *** camera ******** ***** *** PoE. ** ****** ******** a ******* *******, ****** in*** ******* **** ********.

** ****** ***** ******** in **** *****:

Size **********

*** ******** ** ******* are ******** ******* **** some *********** ******, **** as *** ****** * series ** ********* ***** series ******* ***** ****.

IPVM Image

H5 *******

*** ** ** ****** has ** ********* ****** price ** ~$*,***-*,*** *** based ** **** *** typical ******. **** ** similar ** ******** ****** than ***** ******* ** models, **** ** ******** ********* *********** *****-***(**** ~$*,***).

********* ****** ***** ** the ** ***** ****** is ~$***-*** ***, ***** than **** *********** ******, such ** ******* *****-***(~$***-*,*** ******) ******* ********* *****(~$*,*** ******).

Made ** ******

**** ** ******* **** made ** ********'* ******** facilities:

IPVM Image

Test **********

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

  • *.*** *** ****
  • **** ********** *** ******* (15 ***, **** ******** Interval: ***, ********** *******: 14)
  • ******** ** ************ *** enabled
  • **** ******* ***** *** enabled

*** ********* ******** *** software ******** **** **** in **** ****:

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

Note: ********** ****** ******

**** **** ****** *********** analytics *** ***** *******. We **** ** ********* our ******** **********'* ********** ****** *** UMD** ** ******** ******.

Comments (41)

The same results as our tests:

1.- Analytics: impressive

2.- Image quality: Not the best. In fact, for me, not beter than H4A. Noisy and bandwidth hungry.

But, god... THE ANALYTICS

Agree: 5
Disagree: 1
Informative: 3
Unhelpful
Funny

Do you adjust the camera's settings or leave them stock?

Agree
Disagree
Informative
Unhelpful
Funny

We adjust the cameras. But the 4Mpx is noisy in low light. You can adjust the new 3D filter but them you have blurred, dirty images. And de 4K has many WDR issues (camera shake)

Agree
Disagree
Informative: 1
Unhelpful
Funny

Send me a message, please. I have a lot of experience with the H5A cameras.

Agree: 1
Disagree
Informative
Unhelpful
Funny

Send me a message, please.

or don't. but please no more public criticism.

Agree
Disagree
Informative
Unhelpful: 4
Funny: 1

Any news on the cost? Also what is the warranty on it? Great review.

Agree
Disagree
Informative
Unhelpful
Funny

5 year warranty on the H5A and H5SL cameras.

H5A Camera Line | Avigilon

Agree: 4
Disagree
Informative: 6
Unhelpful
Funny

Pricing on the 4K H5A is ~$1,300-1,400 USD estimated street price and ~$800-900 on the 1080p. We've added a section on that to the report.

Agree
Disagree
Informative: 4
Unhelpful
Funny

The price is similar. The same price in 2Mpx, obviously higher in 4Mpx than the old 3Mpx, but similar in all the flavours (2, 4, 5, 6 & 8Mpx). The only crazy thing is the lens options...

Warranty: 5 years :)

Agree: 1
Disagree
Informative
Unhelpful
Funny

Will the H5A cameras integrate directly with SureView for Analytics reporting, allowing you to bypass the VMS? I'm thinking something like NX Witness for recording but the Camera Side Analytics for monitoring?

Agree
Disagree
Informative
Unhelpful
Funny

If/when Avigilon adds H5A ES cameras then yes they can connect directly to SureView. You can also use ES recorders to do the same.

Agree
Disagree
Informative
Unhelpful
Funny

Quick question, does Avigilon offer an SDK/API for third party devs? Wondering why no one integrates with their analytics. (Too lazy to Google)

Agree
Disagree: 1
Informative
Unhelpful
Funny

I think that the SDK will finish with ACC6 and there is no SDK for ACC7. In fact the acc6 sdk works well with ACC7 (but without access to the new ACC7 features)

Agree
Disagree
Informative
Unhelpful
Funny

SDK is going away. ACC7 uses the Web Endpoint API

Agree: 3
Disagree
Informative: 1
Unhelpful
Funny

Jon, there is a restful api/sdk available and those cameras use ONVIF S and T profile. Milestone and Genetec should be compatible in their next releases for use of the analytics.

Agree: 2
Disagree
Informative
Unhelpful
Funny

Hard work pays off in the future.

Laziness pays off right now.

Incoming ONVIF analytics!(not really).

Agree
Disagree
Informative
Unhelpful
Funny: 1

H5A (and H4A) analytics are working with the current release of Genetec (5.8.1.0) and Milestone (2019R2 Device pack 10.3a). I have tested both. Camera generated analytic events are recognized and can trigger actions, alarms and rules. Milestone utilizes ONVIF analytics.

Agree: 2
Disagree
Informative: 4
Unhelpful
Funny

I think you are missing one detail, you compare the size of the han, hik and Avigilon.

I feel you forget to notice the difference in strength of the hinge between the 3 devices, the Avigilon is clearly a better choice when you go industrial.

Agree
Disagree
Informative
Unhelpful
Funny

Stefaan, I am not seeing a material difference between all three, they all are solid and immovable after tightening.

Agree
Disagree
Informative
Unhelpful
Funny

Hi, does anybody have some comparison between low light performance of the H5A 2MP box version (80mm lens) and recent super low light cameras, especially Axis Q1645? Thanks!

Agree
Disagree
Informative
Unhelpful
Funny

Hi Sebastian, we haven't tested the H5A box camera or the Q1645. I'm assuming when you say "80mm lens" you mean the 4.7 – 84.6 mm option?

But we have tested the 2MP bullet H5A and the Axis P3245-LVE. They are not exactly similar but should be in same ball park. Based on our tests, the H5A line is a bit noisier and is prone to some DNR blur. For example, here's the 2MP H5A with the subject walking toward it at ~20':

He's still recognizable to us, but not clear.

The P3245 was less noisy and clearer, though still blurred slightly (different subject, same scene and light level).

I would expect the relative performance of the Avigilon box model and Axis 16 to be in that same range, though not exact.

Curious question: Why the very long lens?

Agree
Disagree
Informative: 1
Unhelpful
Funny

Hi Ethan,

thank you for giving these details! Good hint.

Meant exactly this model (4.7 – 84.6 mm).

Will have to replace some indoor camera looking right through a big city hall.

Agree
Disagree
Informative
Unhelpful
Funny

Update -

Added Genetec to the VMSes that support integration with Avigilon H5 analytics.

Agree
Disagree
Informative
Unhelpful
Funny

Did you use the "Teach by example" or "Self-learning" functions before testing? Or is this the analytics straight out of the box?

Agree
Disagree
Informative
Unhelpful
Funny

By default, Self Learning is enabled on Avigilon cameras. Teach by example is used after the fact if you are having false positives.

Agree
Disagree
Informative: 1
Unhelpful
Funny

H5A starts learning as soon as you install it. Learning is quick enough that I've never needed to use teach by example. Maybe in tough environments with critters and such.

Agree
Disagree
Informative: 1
Unhelpful
Funny

As the others said, self learning is enabled by default, every time the camera is moved we would reset and let it get to 100% before testing. If there are any other false alarms or misclassifications after self learning is at 100% we would then use teach by example.

Agree
Disagree
Informative: 1
Unhelpful
Funny

I understood that in order to fully use Avigilon self learning you had to have an Avigilon VMS. Will the self learning now work with Genetec (no special 'tools' required)?

I go back to VideoIQ (Avigilon purchased) video analytics and all of the analytics configuration with VideoIQ including self learning could be done at the camera independent of VMS.

Agree
Disagree
Informative
Unhelpful
Funny

All learning is still done in the camera. The analytics are then sent (ONVIF) to whatever VMS it is connected to.

Agree
Disagree
Informative
Unhelpful
Funny

We contacted Avigilon directly after having issues. We were told by Avigilon a month or two ago that without an Avigilon VMS, there is now way to adjust / configure the analytics (and we were using current release of Genetec VMS with a new H5A).

Back to the VideoIQ products (before purchased by Avigilon); I could configure the analytics and self-learning at the camera independently of the VMS. I was hoping for this kind of ability again when Avigilon / Motorola 'opened' their products to others. Currently, we cannot do that as I understood from Avigilon - above - that you must have their VMS in order to configure / adjust video analytics and self-learning.

Agree
Disagree
Informative
Unhelpful
Funny

Are you using the Avigilon Camera Configuration Tool?

Agree
Disagree
Informative
Unhelpful
Funny

The Avigilon Camera Configuration Tool only allows configuration of the analytics at a base level. It does not however allow participation in the learning process. For instance, if a bear is detected as a human, it doesn't give the opportunity to help it learn by asking if the detection is correct and giving the option to answer yes or no. This is done in the Avigilon VMS only.

Unfortunately what this means is that a third-party VMS will be inundated with events from either "unknown" objects or those that are falsely categorized. If you don't have access to the Avigilon VMS and this is happening, then you are out of luck.

Agree
Disagree
Informative: 1
Unhelpful
Funny

Yes, I agree to teach by example is only in the ACC software as you need to access the recorded video to use it. That being said we have installed many thousands of Avigilon cameras with analytics and we very rarely have to use that feature.

Other then TBE you have access to enable and disable self-learning. There really isn't any other setup features that are missing...

Agree
Disagree
Informative
Unhelpful
Funny

We have several H4 line cameras watching a gated compound. After about a week of learning (we don't know what the status of the learning was because the tool doesn't state progress), some vehicles that drove in through the gate and parked near the building were not properly identified until they had already parked and the individual was walking toward the facility. The vehicle was in fact identified as an unknown object which is something we're not monitoring. This means that in a dark-screen monitoring scenario, this individual could potentially not be detected until they are well within the area.

Agree
Disagree
Informative: 1
Unhelpful
Funny

My best advice is when you install the cameras have someone walk around in the FOV for 5min so the camera can learn what the object is. If you install a camera and it doesn't see vehicles or humans it will give you false positive. Keep in mind H4A is good and H5A is very good.

I haven't tested Avigilon cameras in other VMS as there is no way they are going to be as closely integrated with the VMS software as they will be with ACC so we stick with ACC. This goes for any camera/VMS manufacturer, not just Avigilon.

Agree
Disagree
Informative
Unhelpful
Funny

So that eliminates all Bear Human colonies. I can live with that.

Agree
Disagree
Informative
Unhelpful
Funny

So having had exceptional results in some applications with VideoIQ products before the company was purchased by Avigilon (and at least some VideoIQ products discontinued, etc.); I was hopeful when I saw the IPVM article 'Avigilon Opens Up Analytics And Cameras' (6 Nov. 2018).

It appears now that in practice little has changed. Unless you are using Avigilon VMS; you cannot access full configuration of analytics or self-learning as it was possible to do with VideoIQ products.

Agree
Disagree
Informative
Unhelpful
Funny

Have you tried ES cameras from Avigilon? They are most like VideoIQ and have the server/storage on the camera. They should be able to do TBE with 3rd party VMS but I have not tested them with 3rd party VMS solutions.

Agree
Disagree
Informative
Unhelpful
Funny

Update -

Added Milestone to the VMSes that support integration with Avigilon H5 analytics.

Agree
Disagree
Informative
Unhelpful
Funny

Please update the H.265 support for Milestone in the table as can be seen on their supported camera page. Thanks.

Agree
Disagree
Informative
Unhelpful
Funny

Update: New Firmware version 7.6 has addressed the annoying issue for numerous events on parked vehicles. We are still seeing some of these issues, but it has improved. Hopefully, future firmware releases will improve these false events even more.

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

This article is part of IPVM's 7,199 reports and 959 tests and is only available to subscribers. To get a one-time preview of our work, enter your work email to access the full article.

Already a subscriber? Login here | Join now
Loading Related Reports