Hanwha WiseDetector Train Your Own Analytics Tested

Avatar
Rob Kilpatrick
Published Apr 30, 2024 13:04 PM

Hanwha's WiseDetector train-your-own analytics, if it works well, would deliver advanced analytics that many have tried but have yet to do well, from our previous testing. But how well does Hanwha's work?

IPVM Image

We tested WiseDetector, a free "machine-learning-based feature" that is available on P-Series cameras that we bought, examining:

  • How effective is it at detecting objects once they are trained?
  • How many training images does it need for accurate detection?
  • How accurate is it in scenes with depth?
  • How accurate is it with objects that change shape?
  • Does it detect multiples of an object after training?
  • How well does it detect the trained objects in different colors?
  • Is training done on-site or in the cloud?
  • How does it compare to other offerings in this segment?

Pricing *** ************

************** "*******-****" *** ** **** ********* with *** *-****** *************. *** ************ ****-** **** ** a ******* ******** ******* *** *** requirements *** ****** ***** ** ****** training **** ***** ** ***** *****.

WiseDetector ********

************ ******** * ******** ** ***** the ********* ***** ************ ****** ******* Device *******. **** *** ******** ** done ******* ** *** ********, *** trained ********* ** ******** ** *** cameras. *****, ** ******* ** ******** of ************ *** ***** ******** ** an ******.

Executive *******

** *** *******, ************ ************ ********** and ******* ** ***** **** ******** PPF, ****** *****/**********, *** *****, **** as * *** ****, ****** ******, or * ********** ****** ** * wall/shelf, ** ****** **** ******* *****. However, ** ******** ******* **** ****** that *********** *** *************:

  • ****** **** ******* ******, **** ** a **** ** *********, ******* ********. WiseDetector ** ******** ** ******** **** objects ** * ***** ***. ********** in *** ******** ** ****** ******* or ******* ******** *****.
  • ** ****** ** ********* ******** ******* of *** **** ***** *** ********* colors, **** ** ***** *****, *****, or ******.
  • ******** ** ******* ** * ****** bounding *** ****. *** **** **** exceeds **** *** ****** ******** **** be ************ ******** ** ******* ***** positives, *** *******, * ****** **** is ******* **** ** ** ********, cannot ** ******** ******* **** ** is ** *** ****.
  • ************ *** ****** *********** **** **** five ********* *******, **** ** ****** without *****, **** ** *** **** of * ******* **** ** ******** people ** ****** *****.
  • *** ****** **** ******** ******** **** one ***** ** * ****, ***** makes *** ******* **** *** *******.
  • ***** *** ********** ******** *** ****** web ******** *********, ************ **** *** integrate **** ******* ****.

************ *** ** ***** ********* ** humans ** *** *****, ** ***** we *** ********* ****** *******. *******, ** *** ******* ******, such ** ******* ******** ** *** object **** *** ***** ** ******* and * **** ** ******** **** there ** ***** ** *** *****. WiseDetector ************* ****************'* "******** ***********," ***** ******** **** ****** *** had **** ****** *********** ***** ********, even ** ******* **** ******** *** same *** ** *** ******* ******.

** ********* ****** ******* ***** **** questions ********* ******* *** ******** ** feedback. ** **** ****** **/**** **** respond.

Improvement **** ***** ****** ***** **** *** *********, ****** ****** *******

************ *** ** *********** **** **** train-your-own ********* **** ** **** ******, such ******* ****** *******. **** ******* ** * ********* object, ************ *** *** ********* ** false ********* ** ****** ** *** scene, ******* ** ******, ** ***** objects ***** *****, ***** *** * big ***** **** ***** ****** *******.

*******, ************ *** *** ******* *** issue **** ***** ** *** ***** or ******* **** ****** ***********/****** *****, a ******* ***** ** ******** **** Bosch ****** ******* *******'* ******** ***********.

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

**** ******* ***** **** *** ************ has ***** *********** *********** ******* ******* or ******* ** * ***** **** no ***** *******, **** ** ***** on * ****, ******* ** * shelf, ***.

***** ************ *** ** ******* *** objects ****** ******* * *****, *** issues **** ***** *** **** **** its ********* *********** ********** *** ********** these *******.

Requires ******** *** ** *** ******

****** *******, ************ ******** *** *** allow ** ** ****** *** ******** box *** ******* ** ******* ********* in * ***** **** ~**' ** depth, ********* * ******** *** ** the *******. **** ** ****** *** moved ******* **** ** ****** *** camera **** *** ******* ********, *** WiseDetector ******** ********* *** ****** ** getting ****** ** *******. **** ** object *** ***** ******, ************ ****** multiple ******** ***** **** *** *******, and **** *** ****** *** ****** further ****, ** ****** *** ******.

Issues **** ******* **** ****** *********** / ****** *****

************ ******** ******* ** ** ** the **** *********** ** ***** ********, reducing ******** *** *** ********* ******** when ******** ** ******* *********** ****** our *******. ************ ******** *** *********** to * ******-**** ******** *** ***** on *** ***** ******** *****. ******* such ** * ****** **** *** upright ** **** ** ******, ** carried ******** ***** *** ** ********** trained **.

***** **** ***** ******* **** ********* orientations *** ********** ********* **** ************. However, ****** *** *******, *** ************ did *** ***** *** ******** ******** box ****** ****** *** ******* ** sideways *******.

Cannot ****** ******* ******* ** ********* ******

******* ******* **** ********* ******, **** as * ***** ****, **** ** be ******* ** ******* ****** ** detect ****, ** ************ *** *** detect *** **** ****** ****** ********* colors.

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

** *** *******, ***** ******* ******** of ~** - ** ****** ** each ***** ** *** ******, ********* many ******* ******* ** *****, ****, and ***** ** *** ******* ****** was **********. ************ ***** ********** ****** bounding ***** ******* ***** ** **** objects ** **** *** ********* ******* in *** ***** ***** ******** ** the ***** ****.

Training **** *** *******

** *** *****, ************'* ******** ******* was **** *** *******. *** ****** required ** ***** ~** ****** ** detect ******* **********, *** **** ***** had ** ** ******* ************. ***** must ** ******* **** ********* ******, so ****** *** **** *** ******** the ***** ********** *** ******** ****.

** *******, ******** ** ****** ** 20 ******, *** ******* *** ********** detection, **** ~** *******. **** ******** training ** *** ************ *** *** moving ** ****** *** ******'* *****. When ******** ** **** **** ~** images, ** ******** *********** ****** ** the *********, *** ***** ****** ** people ** ***** ******* **** ****.

Training ******** ******* ** ****** ** ****** *******

** *** *******, ******** ******** ******** on *** **** ****** *** ****** was ******* **.

  • ~*-** ****** ** *** ****** ***** cause ***** ********* ** ***** ******* and ********** ******** *****.
  • ~**-** ****** **** ********** *** ********, though ******** ***** ***** **** *** the ****** ** ********* ******.
  • ~**- ** *** **** ** *** as *** **** ****** ** ****** for ********; **** ******* *** ******** boxes ** ***** *** ****** ******** and *** ***** ***** ** ******.
  • >** ******, ** *** *******, ***** result ** ************ *** ***** ********* on ****** ** *** *****.

Object *********** *** ******** ****** ********* ******/*******

** ***** **** ************ *** ****** with ***** *******, *** ******* ***** not ** ******* *** **** ******** across ******** *******. ******** ******* **** be ******* ************ ** *** ****** in ***** ***** ** ****** ** object ********.

Limited ** *** ******* *** ******

************ ** ******* ** *** ******* per ******, ***** *** ****** *********** in ***** **** ******** ******** ******* it *** ** ***** *** ****** in *** ***** ******** ** ****** the ****** ** ***** *********.

IPVM Image

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

************ *** *** ********** **** ******* Wave *** *** ****** ********* **** during *** ****.

IPVM Image

Versions ****

*** ********* ******** **** **** ****** testing.

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