Axis AI-Based Object Analytics Tested

By Rob Kilpatrick, Published Jan 28, 2021, 10:13am EST

Axis has finally released deep learning analytics in their Q1615 Mk III with Axis Object Analytics, but does it live up to their claims of "smarter surveillance"?

IPVM Image

To find out, we bought and tested the deep learning Axis Q1615 Mk III and tested Object Analytics on the machine learning enabled P1375-E, examining:

  • How does performance differ between deep learning and machine learning models?
  • How do the cameras handle common outdoor false alert sources like animals, foliage, and shadows?
  • How well do they handle indoor false alerts from shadows, lights, noise, etc.?
  • Can both analytics detect people running, crawling, or obstructed?
  • Are people and vehicles properly classified?
  • How far can both analytics detect people?
  • How does image quality compare?

****: **** **** *** delayed ******* ********* ************ **** ******** ***** Mk *** ** **** 2020 *** ******* ** shortly **********, *** **** learning ********* *** ** was **** ******** **** month, ******* ****.

*******

**** ****** ********* ** the **** ******** ******** (DLPU) ***** ** *** was **** ****** ******* while *** ******* ******** (MLPU) ***** *****-* *** some ************** ********.

*** ***** ** *** avoided ***** ****** **** common ******* ******* *** out, ********* *******, *******, shadows, ***** *******, *** others, *** ********** ******** subjects **** *******, ********, and **********. ******* *************** in *** **** ******** model **** ********, ******** classifying ****, ******, *****, and *****.

*******, ** *** ******* learning (****) ****** *****-*, Axis ****** ********* ****** to ****** ******** ********, occasionally ************* ****** ** vehicles *** **** *****, and ******** ***** ****** on ***** *******, *********** them ** ******. *******, it *** *** ******** rain, ****, *******, ******, etc ** *******.

**** ********* ******** ~** PPF *** ******** ***** detection, ******** ** *** tested ********** **** ** Avigilon *** ** ***** IVA.

****** ******** ****** ********** with *******, *********, *** Nx ******* ** *** tests, *** *** *** work ** ******** ******* Center ** *****.

Vs. ********, *****, ******

** ***** ** ******** and ***** ***** ***********, Axis ****** ********* ** deep ******** ****** **** the ***** ** *** is *********** ******* ********** ** **** tests, **** **********’* ***,***** ***,***** *********+, ********’* **, **** ****** ***** alert **********, ******** *********, and ******* *****/*** ************.

*******, ** ****, ***** manufacturers *** ***** ******** which *** **** ***, such ** ******** ****** metadata (*****, ******** ****, vehicle ****, ***.) ***** can ** ****** ** and ******** ** *****. Axis ****** ********* ***** only ***** ***** **** (on/off) ** *****, **** no ****** ** ****** metadata.

Axis **** ******** **. ******* ******** ******

**** ****** ********* ********* only ******** * ******** of ****' ******* ******* and *** ***-********** **** learning ******* ** **** 3 ****** **** ******* 30 ********** *** ******* learning ******.

IPVM Image

*** *** ** ************ in **.*/**.* ********, **** no ******* ***.

*** ** ***** ****** use****' *** ******-* ***, *** **** ******** models ******* * ****** processor (****). **** *** declined ** ***** **** chip ** **** *** the ****.

*** ********* ***** **** can ** ***** ****. **** **** **** they **** ** ********* more **** ******** ****** in ****, ********* *, P, *** * ****** cameras.

Deep ********: ** ****** ** ****** *******, ********, ********** ** *****

** * **** ** testing, *** **** ****** did *** **** ****** running ** ******** ******* the *****.

IPVM Image

Machine ******** ****** ****** ****** ********

*******, *** **** (******* learning ********** ****) ****** missed ****** ******** ******* the *****, ***** ****:

IPVM Image

**** ********* **** **** was ******** ******** *** machine ******** ******, *** that **** *** ******* into ********* ********* ** future ********:

**** ** ******** ******** for **** ****** (****). Crawling ** *** ** specifications ***** *** ***** it ** ***** ** is **** ******* *** upright *******. **** ** also ********* ** *** documentations. *******, ** *** looking **** *** ** can ******* **** ****** forward.

Obstructed ****** ********

**** **** ******** *** machine ******** ****** ********** detection ****** **** ************* obstructed:

IPVM Image

MLPU ****** ********** ***** ** *******, ******* ** *****

** **** ********* *** MLPU ****** ********** ****** as ********, ***** ***** the **** ******** *** below:

IPVM Image

*** ** **** ***** vehicles **** ********** ** humans, ** ****** *** cases ** *****, ***** here:

IPVM Image

**** ************ **** **** risk ** ***************** *** something **** **** ***** of, ** ****:

*** **** ** ***************** is ********* ** *** aware ** *** *** continuously ******* ** *******. AXIS ****** ********* ******* on **** ******* **** as *** **** *****/-* is ***** ** * more ***** ****** ***** compared ** *** **** cameras, *** **** ** something **** ************* *** occur ************. ********** ** the ****** ** ******* or ******** ***** ** some ***** ****** ***** makes *** ******* **** challenging ** ******** *********. We **** **** ***** support *** *** ***** on ******* ******** **** more **** ** **** models. *** **** ********* scenes **** ****** ************ we ********* **** ** our **** *******.

No ***** ****** ** ****** ****** *******

******* ****** ********* ***** alerts ** ****** ******* such ** ***** ***** changes, ***** ****:

IPVM Image

** ******* *** ****** cast ** *** *****:

IPVM Image

Deep ******** - ** ***** ****** ** ****** ******* *******

****** ******* ******* ** false ****** **** ** swaying ******* *** *** shadows ******** **** ******* by **** *** **** and **** ******.

IPVM Image

*** **** ******** ****** also *** *** ***** alert ** *** ******* in *** *****.

IPVM Image

****** ** *** ********, the ****** ** * deer ******* *** ****** classified ** * ******. This ******** **** **** in *******, ******* ***** deer ***** ** *** FOV *** ~* *****.

IPVM Image

Machine ******** - ***** ****** ** *******

** *** *****, *** machine ******** ****** ******* classified **** ** ******, shown *****.

IPVM Image

No ***** *************** ** ***** ****/****

* ****** ****** ** false ****** *** **** other *********, ***** **** and **** **** ******* by **** **** ******, with ****** *** ******** accurately ********.

IPVM Image

Vehicle **** *************** ********* ********

** **** ******** *******, Axis ****** ********* ******** vehicle ****, ********* ***, truck, ***, *** ****, while ******* ******** ****** provide **** *** ******* "vehicle" **************.

** *** *******, *** deep ******** ****** ********* correctly ********** ******* ****, such ** *** *** below:

IPVM Image

"*****" ************** *********** ***** large ****** *** **** in *** *******, ********** shown ****.

IPVM Image

******* *** **** **** classified ** **** ** our *******, ********* **** Axis.

IPVM Image

****** *******, ** ******** a ***** ****** (of ***** ****** ***** were ********** ** ****, but **** *** **** rare, *********** ** ******** hundreds ** ******** ** our *****.

IPVM Image

**** ********:

** ** ****** *** shape ** *** ****** and *** *** **** itself **** ** *** base *** ******* **************. As *** ******* ***** is *** * ***** definition ** ***´* *** pickups *** **** *** also **** ******* ********* models. **** ******** **** the **** **** ******* may **** *********** **** different ******, ***** ** challenging *** *** *********** to **** * ********** behavior ** *** *****. In ******* ** ***** expect ***´* *** ******* to ** ********** ** cars *** ***** ** angle ** ********** *** other ******* **** *********, this *** ****.

~11 *** ******** *** ****** *********

****** ******* ~** *** was ****** *** ****** detection ** *** **** and ~** *** *** needed ** *** ****.

IPVM Image

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

****** ********* ****** ********** with *******, *********, *** Nx ******* ** *** testing, *** *** ******** or *****.

IPVM Image

** ********* ***** (*.*. Genetec, ***** *****), ***** may ****** *** ******* event ******* *** ***, or ******** ***** (*.*., just ******, **** ********, etc.).

IPVM Image

**** **** **** ****** were ****, *.*., ****** detected ** ***** ********, but *** ****** ******** as ** ****** **** several *********** ******* **** as****************.

Burnt-In ******** *****

**** ** *** ***** tested ******** ******** *** information *** ********, ******. However, **** ****** ********* allows ******** ***** ** be "***** **" ** one ** *** ******'* streams, ** ** *** be ******* ** *** VMS.

IPVM Image

Simple *****

***** ** ******* ** other *********, *********** **** requiring ***** ** ****** what ******* ***** **** to ****** (******/********) *** the **** ** ****** in. **** ****** ** each **** ** * "Scenario" (***** *****), *** users *** **** ** to * ********* *** camera.

**** ******** ******** ******** from ***** ******** ********* such ** ***** ****, swaying *******, ***** *******, or *** ******* ** set *** ****** ** different ******* ** *** scene. *******, ** ***** in *** ******* **** these *** *** ****** performance ** ****** *** DLPU ** **** *******.

IPVM Image

Q1615 ** *** ******* ***********: ****** *** *** ***** *** *****

*** **** ***** ** III's *** ******* **** strong, ******* ** **** cameras ** *** **** Q-Series ****.

IPVM Image

***** ****, ** *********** the ***** ***** **** competitive ****** **** ****** and *********.

IPVM Image

************, ****** ** ***** at ~* *** *** very ******, **** ****** and ***** ******* ****** than ***** ****** ******, below:

IPVM Image

*******, ** **** *** light (~*.* ***), *** Q1615 ** ***, ****** facial ******* **** ** identify *** ****** ******* darkened, ***** ****** *** Hikvision ****** *** ********, though **** **** **** noise, ******** *******.

IPVM Image

*******, *** **********, **** light ******* ******** ****** on *** ******* ******, with *** ******* *** test ***** ******* ******* and *******.

IPVM Image

Versions ****

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

  • **** ***** ** ***: 10.3.0
  • **** *****-*: **.*.*

Comments (7)

*'* ** ********** ** know ******* **** ****** deep ******** ***** *******? Looking ** ***** ******* it ******* **** *** P3255-LVE ** * **** camera **** **** ******** but **** * ********* resolution. ** ********* ******* dome ******* ** *** 3-8 ** *****.

Agree
Disagree
Informative
Unhelpful
Funny

* ***** * **** to ******** **** *** article ******* ** ******** next ****. *** ******* covered **** :)

Agree
Disagree
Informative
Unhelpful: 1
Funny: 3

**, ** ** *** best. **** ****'* ******* any ****** ** ******* of ***** **** ******** models ** *** ******* differences.

*** ***** ** ***** $100 **** **** *** non-deep ******** *****, ** that *** **** ** some ****. ** **** be **** *********** **** they ******* * ****** models. ******* ** *** how ****'** ******.

Agree
Disagree
Informative
Unhelpful
Funny

*'* ******* ** **** it ***. ** ************ with ****** ********* *** a *** ***. *'* to *** ***** ***** if * **** ******* by ***'* ********** ** a ***, *'** ** impressed.

Agree
Disagree
Informative
Unhelpful
Funny

**** **-****** ********* **** you **** ******* **** set **** ************ **** low?

Agree
Disagree
Informative
Unhelpful
Funny

** ***** ** *********** to **** **** **** will ****** *** **** cameras **** ****. **** anyone **** ****?

Agree
Disagree
Informative
Unhelpful
Funny

**** ***********,***** ***

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

This article is part of IPVM's 6,891 reports, 921 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