IronYun AI Analytics Tested

By Rob Kilpatrick, Published Feb 17, 2020, 11:43am EST

Taiwan / US startup IronYun has raised tens of millions for its "mission to be the leading Artificial Intelligence, big data video software as a service (VSaaS) solutions provider." But how good is the product itself?

We tested the product for two weeks including detection of people, backpacks, handbags, guns, animals plus color detection and facial recognition to see what strengths and weaknesses the system had.

*******

******* ****** **** ****** person ********* *********, ********* backpack *** ****** *********, along **** ***** **** rec ********** ** **** lit ******, **** ** harsh ******.

*******, ** ******** **** some ****************** ** ********* on ******* ***** *** on ***** ******* ***** may ** ** ***** in **** ************.

*******, ***** **** ******** VMSes *** **** *******'* lack ** *** ******/***** integration ** ***** ** operators **** ***** *** use *** ******** *** to *** ***** ********.

Key ********

***** ** *** ***** of *******, ******* ******* several *********** *********:

  • ****** ****** *********:****** ********* *** *** of *******'* ********* *********, detecting ****** ** ****** walking *****, *******, ********, and **********. ****** **** even ******** ******* ******* or **** ****** *** by **********.
  • ***** ******** *** ******* Detection:************, ******* ********** ********** people **** **** ********* or ******* ****, * feature ***** *** *** work ************ ** ***** analytics.
  • *** ********* ********:******* *** **** **** were ******** ********** **** they ***** ** ******** in *******, *.*., ****** or **** ******* ** shooting ******. ********* *** not ********** **** **** were **** ** *** subject's **** ** **** facing *** ****** **** gun ******.
  • ***** ****** *********** ** Well *** ******:** *** *****, *******'* face *** ********** ********** subjects ** **** *** daytime ******, **** **** using ****** ********* (~**°) and ****** ** ********* (~45°).

*******, ***** **** ******* negatives:

  • ***** ****** ** ***** Animals:***** ******* **** ** deer ********* ********* ***** person ******.
  • ***** *************** ** ****:******* ********** ********* ** cameras' ***** ** ******** object *****, ********* ******, airplanes, ****, *** ****.
  • ****** **** ********* **********:******** ** *** ****** edge ** *******' *** (with ***** ***** **** out ** *****) **** often ********** ** *******. People ******** ******* ** full **** ** *** camera **** ********** ****** or ********** ** ****.
  • ******* ***/** ****** *********** Performance:*******'* **** *** ********* failed ** ******* ****** against ****** *********. ********** in ** ****** *** reduced ** ~** ******, causing ******, ** ****.
  • ** ****** *** *******:*******'* ****** (*********, **** rec, ***.) *** ****** are *** ********** ** third ***** ***. ****** events *** ** **** to ***** ***** ****, but **** ************* **** be ********* ********. ******* does ********* ** ***** to ******** ***** (**** and ********) *** ***** is ** ***** ***********.

*******

******* *** ** ********* street ***** ** ~$***-*** USD *** ******* *** live ********* (********* *********, facial ***********, ******** ******, etc.) *** ~$***-*** *** video ******, *** *******.

***** ****** *** ************* lower **** **** ***********, such ***********, ***** ***** *** over $*,*** *** *** channel, ***********, ***** *** * $20,000 ******* ***, ** a *** ******* ******* at $*** *** *******.

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

** *** *****, ****** detection *** *** ** IronYun's ******* *********, ********* people ** **** ********* scenes, **** ** **** washed *** ** *** headlights:

** **** ** **** partially ********, **** ** outside ** *** ****** below.

** ****** **** ****** when ******* ** ********, either.

Accurate ******* *** ******** *********

************, *******'* ******** *** handbag ********* *** **** accurate **** ***** ********* we **** ****** ***** claim ** ****** ***** objects. ****** ******** *** or *** ***** ** both **** ******** ********.

Person **** *** ********

****** *******, ****** ******* on *** ****** ** crouching **** *** ********** detected *** ********* ********** as *******. ************, ******** were ******** ** ******* when ***** ****** **** was *** ** *** scene *** *** *** half ******** ** *** bottom ** *** ***, shown ****:

Gun ********* ******** ** **** *** ** *****

*** ********* *** ******** during ******* **** *** gun *** *******, *.*., when **** ** ** when *** ******* *** in *** ******** ******. Detection *** *** ******** when ****** ****** *** camera ** *** *** could *** ** **************.

****** **** ******** **** held **** ** *** subject's ****, *** *** as ******** ** **** held **. ******** **** not ******** **** **** down.

False *************** ** *********

********* ** *** **** of * ****** ***** be ******** *** **** of *** ********* *************** that ******* *******, ********* people:

** **** ** *********, bats, *** **** ***** classifications.

***** ****** ***** ** improved, *** *** *******, via ******** *******.

False ****** ** ***** *******

***** ******* ******* ******* the ***** **** ********** classified ** ******. ********** minimum ****** ********** ******* the ********* **** ******** but *** *** *** rid ** *** ***** alerts.

Poor **** *********** *********** ** *** *** **

****** ******* **** *********** performance ******* ** *** scene, ********** ******* ****** walking *******.

** ** ** *****, confidence ********* ******* ~** points ********* ** ****** even ** ***** *********.

Color ******** ** **** *****

********* ****** ** ******** color ***** ******** ** the ***** ******** ** either *** *** ** bottom ** ********, ***** a ****** *** *** correctly ***** ****** ******* red.

******, **** *** ** search ** **, ****** only **** ** ** white, ****, *** *****.

Alert *****

******* ** ** ***** in ******* ***** **** a *** *******. *****, the ******* ****** ** selected (*********, ****** ***********, fall *********, ***.), **** a **** *****. *******, specific ******* *** ******** (e.g. ****** **. ****** vs. ********).

Weak *** *******

******* **** *** ********* its **** ****** ** search ** *** *** out ** *** ***. Users *** ******* ****** in *** *** ***** HTTP, *** **** **** be ********** ********. ******* forensic ****** ** *** available **** *** ***.

**** **** ***** ** support *** ******* **************, ********* ***** *** Milestone, ** **** ** Hikvision *** ***** ****. These ************ ******* ******* from ******* *** ******** in *** **** *******.

Live ******** **

******* *** *** **** types ** **, **** and ******. *** "****" UI ******** * ******'* live **** ***** **** an ***** **** ** whatever **** ** ***** users *** ******* (********* shown *****). ***** *** click *** ***** ** begin ******** *** *** which ******* **** **********.

*******'* ****** ** ****** filtering ** ******* ** type, *****, *** ***** parameters, ** ******** ******* and ** ******** **** ranges, *** ******* ** the ******* ***** **** used ** *** **** analytic **** ***** *****.

***** ** ** ******** video ** *** ******* UI.

Comments (16)

** **** *** ** age, ***** ****** ** MUCH *****.

*** *** *** *********'* for ********* ******* ** pay * ******* *** get *** **** ***** much ******, ** *** test ******* ****. ****** choice, *****?

****** ******, *** **'* be *********** ** *** a **** ** *** bottom **** **.

****** ******, *** **'* be *********** ** *** a **** ** *** bottom **** **.

****, **** ***** ******* is ****** ******* ** the ***! ** ****** all ******* * ***-******.

🙄**** ** ** ** that *** **** ********* in * **** ***** situation *** ****, **’* worth **. *** ***** expensive ** ******* ** matter **** **** ***, they’re ********* ** * controlled ***********.

*** ***** ********* ** systems ** ****** **** they ***, ****’** ********* in * ********** ***********.

**** ***** ***** *** that *** ***** *** too ****. *** **** point ** **** **** are *** '********* ** a ****-***** *********'?

** ****** ** * 'real-world *********'.

* ** ****** ****** to ********** *** ***** of **** *********. ***** are ****** ** ********* and ** *** ** the ****** *** *****'* also ***** * *** of ********* **** '****-*****' testing.

***** *** ************* ******* alternatives **#*, * ***** it **** ******* ** a **** ** **** basis **** *** **** out ** *** ********.

** *** * **** unit ** **** *** but ** **** **** it *** *** **** refined. ** **** ******* it **** ** ******** system *** *** *** to **** ********** ** via ****.

** *** *** **** itself ** * **, based ** ***** *******. Perhaps ***** *********, *** just ** ** *****, you **** ** ******* the ******(*), *** *** each ****** *** *** up ** $*** *** "live *********" *** ** to ******* $*** *** "video ******"? *** ***** options ** ****** ***** on ******** ****** ****? The ********* **** ****** later ********, ***? ** "non ****"?

** "***** ******" **** mean ***** ****** ****** module (********* ** **+ objects). *** ***** **** not *** ******** ******** to *** ******* ******, it ***** ***** **** an *** ** *** (with * ******** ****** from *******) ** ** there ** * ******** event ** **** **** the **** **** *** NVR/VMS ** *** **** of *** *****. *** analytics *** ********* *** live ****** ** *** camera, *** ********.

** *******, ******* ***** their ****** "** ***" but ** **** *** function **** ** ***, it **** ***** *** snapshots **** ** ***** and **** *** ****** video ** *** ******. To ******** **** ***** from ** ***** ** needs ** ** ****** with ** *** ** VMS (**** * ******** driver **** *******)

* ****** ** ** depth *******. *** ********, I ** ********** ** find **** ***** *** hardware. *** **** ******** does ***** ** *** support ** **** *** the ******** ************ ** run ** ** **** own ******?

** ***,

****** ******* ** *******@*******.***, ** **** ** glad ** ******* ***.

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

**’** * ******** ********** and **** ********* ******** IronYun ** ******* ********** several ******** ********* ** a **** ****** ***** enterprise ***********. *****, ** did ********* ********* ***** alarm ****** ** **** drops. **** ***** *** easily *** *********** ********* and *** **** ********** eliminated. **’** **** ********** results ** *** ******* situations ** *** **** 90 **********.
******, ******* **** ******* an *** **** ********* to **** ******* *** platforms. ***** ******* **** popular ** ********* ** say **** ********** ******* proved ** ** **** of *****. ** **** them ** *+.

****, ******** ** **** first *******. **** ******* did *** **** ** completely ********* *** ***** alerts ** ****? ** found **** **** ********** the ********** ****** ** alarm ** * ****** wouldn't *** *** ** the ***** ******.

**** ** **** **** IronYun. ** ******* *** confidence ********* ** ****** objects ** *.** *** the ******** ******* ***********. I ***** **** ********* used *.** ** *** Person ********** ********* *** that *** ** *** cause ** *** ******** false ******. ** **** seen ****, ****, *** ice ** **** ****** lenses ******* *** ***** alarms *** **** * year.

Read this IPVM report for free.

This article is part of IPVM's 6,803 reports, 913 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