Hanwha AI Object Detection Tested

By Rob Kilpatrick, Published Sep 28, 2020, 01:07pm EDT

Hanwha has added detection and classification of people, cars, clothing style, masks, age, gender, glasses, hats, color and more to their AI cameras. But do these new analytics and search perform in the real world?

IPVM Image

We tested Hanwha's object metadata analytics using their P-Series analytic camera and TNB-9000 8K camera, examining:

  • How accurate is searching by specific colors?
  • Are vehicle classifications accurate?
  • Does it accurately classify age and gender?
  • How does it compare to motion based video search?
  • How does it work in full light/low light/IR?
  • How does it compare to Avigilon Appearance Search?

Executive *******

** **** * **** of *******, ******'* ** produced ******** ***** ******** (person/vehicle, *****, *******, *****, etc.), ****** ** ****** to ****** *** ******** objects ******** ** ****** video ******. *******, *** and ****** *********** *** inaccurate, **** *** ******** as *****, ******** ** adults, *** **** *****.

************, ******'* ******* **** VMS ****** *** ****, with ** ********* ***** of ******** * ******, requiring ***** ** ****** and ***** ******* ****** to ****** *** ****** types, ******, ** ***** specific ***********, ***** ******* and ********* ************ ******* users ** ***** ** choose ***** *******.

Vs ******** ********** ******

********* *** ******** ****** in ******* **** *** more *********** ************'* ********** *********** ***** **** ***** specific ****** *** ********* identifiers. **** ** ******** in ******* *** ********* integrations ***** ***** ***** to ***** ** ****** specific ******* *******. *******, none ** ***** ************ allow ***** ** ****** click ******** ****** ** live ** ******** ***** to ****** * ******, a *** ********* ********* of ********** ******.

******'* ******** ********* *** search ** *** **** the *** ************ ** Appearance ******, ***** ******* hardware ****. *******, ******** analytics *** ********* **** in******'* ** *-****** ** cameras***** ***-**** ******, ***** ********'*** ************ ********* ** * broader ********* ** ******* and ***** ******.

Complex ****** ** ******* ****

******'* ** ******** ** integrated ** ******* ****, with ******** ********* ** the ***** **** ***** tab **** ********, **** their ******** *********** ******. Bounding ***** ****** ****** objects ** **** *** recorded *****, **** **** showing *******/****** ******** *** descriptive ****, ** ****, shown ****:

IPVM Image

*******, **** ***** * simple ***** ** ********* anything ***** **** ***** criteria (******, *******, ***.). In ***** ** ****** colors, ****, *******, *****, etc., ******** **** ******** enter *** ****** ****** into *** ****** ***'* search, *.*. "***: ****", shown *****. ***** ** no *** ** ***** to ****** ***** ***********.

IPVM Image

Genetec *** ********* ******* ******

** ********, ******* *** Milestone **** ******* ****** dialogs ***** ***** ***** to ***** ******** ** clicking, ********* ****** ****, color, ******, *** ***** descriptors, ******* ** *** text ****** **** ** Wave. **** ****** ** more ******* ** **** is ***** ** ********'* Appearance ******.

** ****** *** *********** between **** *** *******/********* in **** *****:

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

** *** *****, ********* upper *** ***** **** clothing ****** *** ******** during *** *** *** in ******** *** ***** (~1-5 ***).

IPVM Image

******** ****** **** **** accurate (***, ******, ******, blue). ****** ******, *.*., brown, ****, *** ***** were ********* ******** *** could ********* ** ******** if ******** ******* ******** areas.

Integrated ** *** *******

** *** **** ***** IR, *** ******** ****** either ***** ** ****, limiting *** ****** ******** to ****** *****, ****, or ***** ***** *** lower **** ********* ******* IR *****.

IPVM Image

**** **** **** ********** is ******* ********* **** color ******, ******* ** Hanwha, **** ********** ********** ****************.

Clothing ****** ********* / ****** ******** ** "****"

******'* ********* ****** ******* subjects ** ******* **** sleeves *** *****, **** when ***** ****** ****** or ****** **** ****. This *********** ** ********* only ***** ** ******* Wave.

IPVM Image

****** **** **** ** a ***** ********** *** will ** ***** ** future ********:

** ** ***, *****/***** clothing **** *** ******* hats *** *** ******* yet. ** ******* ****** long / ***** ** this ******. ** **** be ***** ** *** camera **** ****. ** is ******* ***** **********.

Reliable *******/**** *********

******** ******* ******* *** scene **** ******* ** a **** **** ********** identified ** **** ** our ******* *** **********. Avigilon ********** ********* ** ***** H5 ****** ****** (*** our ****), *** ** ** not ******** ** ********** Search **********, *** **** do *** ****** *******.

IPVM Image

Accurate ******* ****/***** ******

******* ************** *** ********, correctly *********** ****, *****, trucks, ********, *** ***********.

IPVM Image

*******, **** **** ********* identified ** **** **** and ******, ***** ** the ******* *****, * few ******* *****:

IPVM Image

***** ****** *** ******** was **** ******** ****** the ***, ***** *****. Nighttime *********** *** ******** by ** *****, **** vehicles *** ******** ** white/black/grey, ********* *****.

IPVM Image

*******, ** *****, ******* color ****** *** *** same ****** ** *******.

Dumpsters ********** ** ********

******** ******* ****** *** generally ********, ********* *** dumpsters **** ********** ** blue ****:

IPVM Image

Parked ******** ************ ********

******* ** ***** ** analytics ** **** ******, Hanwha's ****** ********* ******** constant, ********** ********* ** parked ********, ********* ** frequent ****-******** ***** ********** as **** ** ******** were ****** ** *** field ** ****.

IPVM Image

*** **** ******** ** this ***** ** ****** clutter, ** ******* *** quickly ***** ** ******* each *****, ****** ********** more *********. ******** *** still ********** ** ***** these ****** ** *** stop ***** ****** ******** from ***** ****/******.

Searching ***/****** **********

**** ********* ***** ********, searching ** *** *** frequently **********, **** ******** almost ****** ********** ** adults.

IPVM Image

************, ****** ****** *** inaccurate, ********* *** *** women *** ***** *** men.

IPVM Image

Camera ******** *****

***** ** ****** ********* for *** *-****** ******* is ******, **** ***** simply ******** *** ****** types **** **** ** see (******, *******, ***.). Object ******** ** **** to ********** ***** *** search.

*******

******'* ****** ******** ** included ** * ****** AI ******* (~$*,***+ *** online) *** ***** ** TNB-9000 (~$*,*** ***, *** including ****). *********** ** Hanwha ******* **** (*** Nx *******) *** ********* requires ** ******* *********. Genetec *** ********* ******* are ********* **** ****** at ** ******.

Versions ****

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

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

Comments (5)

* ** ******* *** same ******** *** * different ******* ** **** and * ** ****** lots ** ******. *** you *** *** ***** detections *** *** ***** info ** **** ******* like *****?

IPVM Image

Agree
Disagree
Informative
Unhelpful
Funny

** *** **** ***** info ** ** ******* version ** **** *** the ***** ******* ***** it *** *** * lot **** *********** **** tracking *******.

IPVM Image

*** **** ***** ********** that ** *** ****** testing **** *** ********* to *** ***** ********* classified ** * **** car.

Agree
Disagree
Informative: 2
Unhelpful
Funny

**** ** *** ****** on ***** **** ***** is **** * ** using. ** ** ********* useless **** ***** *******.

IPVM Image

Agree
Disagree: 1
Informative
Unhelpful
Funny

****** **** ** *** latest ***** *** *** issues *** ***** **** the ***** **** ** the ********.

IPVM Image

Agree: 1
Disagree
Informative: 1
Unhelpful
Funny

********* ********** ** * blue ***.

**** ***** ****** ** mistaken ** ** *** Volvo. ;-)

IPVM Image

Agree
Disagree
Informative
Unhelpful
Funny: 6
Read this IPVM report for free.

This article is part of IPVM's 7,253 reports and 966 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