Hikvision DeepinMind 2019 Test

By Rob Kilpatrick, Published Jun 06, 2019, 11:17am EDT

In 2018, Hikvision's DeepinMind AI NVR performed terribly, recognizing vehicles, animals, and other objects as humans, misclassifying demographics, completely missing humans in the scene and more issues.

free-images

Now, after a year of firmware updates, Hikvision claims significant improvements in human detection and more accurate intrusion performance.

We tested the latest firmware on the iDS9632NXI-I8/16S DeepinMind NVR we bought for our original tests to see how it performs, examining:

  • False alert performance: What did and did not cause false alerts?
  • Intrusion detection accuracy: Did it miss valid people or vehicles in the scene in multiple conditions?
  • Ease of setup: How difficult or time-consuming was setup and optimization?
  • Vs. competitive analytics: How did DeepinView compare to Avigilon's self-learning camera analytic cameras?

*******

***** ** * ***** of ******* *** ********* thousands ** ******, *** DeepinMind *** *** **** positives:

  • ** ****** **********:** ****** *******/******* ** vehicles ******* ******* *** scene **** ****** ****** a ***** ** ******* in ******* ********** ********* day, *****, *** ***** rain.
  • ** ****** ** *******:********** *** *** ***** on *** ******* ** our *****, ********* ******* wires, ********, ** ***** objects.
  • ** ****** ** ******* brush:******* *****/******** *** ******* by ********** ********** *******, a ****** ****** ** false ****** ** ********* *********.

*******, **** ******** ***** suffered **** ******** ****** seen ** *** ******** tests ** **********, *********:

  • **** ** **** ******** alerts:***** ******** ******* **** camera ***** ********* ****** during ***** **** *****, light ** *****. ****** were *** ********* ** raindrops **********, **** ********* on *** ****.
  • *********** ******* ******:*********** ** ****** *** outdoor *****/********** ** ***-****** vehicles ***** ********* ***** positives.
  • ************* ****** *** ********:****** *******, ********** ************ misidentified ****** ** ******** and **** *****.
  • *****/***** ******* ******* ****** alert:******* ****** ** ***** alerts *** ***** *** large ******* ******* ******* the *****, *** ****** would ******** ** ********** as * *****.

**** **** ** ********* to ****** **** ******** as ********* ** ********* technical *******, **** ** threshold (******* **** ****** is ** *****) *** minimum *** ******* **** settings, *** ******* ** these ******* ******** * material ****** ** ***** alerts.

Vs. ********* ********* ********* / ****** ***

******* *** *********** ***** alert ******, ********** ***** outperforms *********'* ******* **-******/*** intrusion ********* ********* (*** *** ****), ***** ********** ******* on ******** *******, **** as *****, *******, *******, light *******, *** **** specs ** **** ** some *****. ** ***** where ******* *** **** will *** ** ******, such ** ******** ************, DeepinMind *** ******* ******** detection ***** ******** ********* may ***.

************, ******** ** ******** VMD, *********** ** *********** better, ** **** ****** side *** ***** ****** alarm ** *** ***** change, ********* *** *** limited ** *** ****** false ***** ******* *****, without ************** ******* ******, vehicles, ** ***** *******.

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

** *** *****,*********'* ************** ******** ******** ****** offered ****** ***** ***** performance **** *** ********** NVR, ******** ****** **** rain ** ******** ** shadows *** *****. ************, false ****** **** ***** sources **** **** ******** in *** ********** ******.

**** *********** ********** ** likely *** ** ********** used, ** *** ********** NVR **** * ******, older ********** **** ******** chip *** *** ********, while ********** ******* *** a **** ******** **** for * ****** ******* of *****.

Vs. ******** ****-******** *********

** *** *****, ********** was ********* ********* ** false ***** ******* ***** Avigilon's ** ****-******** ******** cameras ********. ** *** tests, ** ******** ****** in *** **** ****** was *** ********* ** any ******* ****** ******* days ** *******, *** light *******, *******, *******, etc.

Little ********** ** ****** ***********

** *** *****, *** DeepinMind *** ********** ********* regardless ** **** ****** of ********* ****** *** used *** *******. ***** were ********* **** *, 4, *, *** * series ******, **** *********** effectively ******* ******* ****.

Updated ******** ******

**** **** *** ********* using *** ******** ******* of ********** *** ********. However, ******* ** ***** America, ********* *** ******** a****** ********** ** *** DeepinMind *** ******(***** *** **** **** numbers **** * * suffix).

** ***** ********* ***** performance *********** ******* *** first *** ****** ***********, as **** ** ************ of * ******, ** which **** *********:

********* *** *** *** generation, ** *** ******* accuracy ******* ** ***** of ******** **** *** generation ****** ************** ****** rather **** ***** ******...*** 2nd ********** *******, ** plan ** ***** **** in ****** ** *********.

Simple *************

************* ** ********** ** simple, **** ********* ***** to ** ******* ** the *** ****** ***** settings *** "****** ***** Smart ********" ******* ** the ********* ******* ** the ***.

***** ******** **** ** threshold (**** ** ****** in *****) *** *********** can ** ******* *** did *** **** ** impact ** ********* ** our *****.

Rejected ***** ****** ** ***** *** *******

********** *** *** ***** on ******* *****/***** ** the ***** ** **** during *******, **** ** the ***** ** *** left *** ***** *****, one ** *** **** common ***** ****** ** previous ********* *****. ************, shadows **** *****, *****, or ***** ******* **** rejected, * ****** ****** which ********* ***** ********* in **** *****. **** are ***** ***** ******* in *** ****.

No-Alarms-On-Blowing-Branches,-Foliage,-or-Shadows

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

********** *** ******* ****** with **** ** ******** on *** **** *** heavy ****, ** ***** that******** ****-******** *********,***** ***, *** ************** **************** *** *** ****.

Rain-On-Dome-Causes-False-Alerts

** ******** ** **** on *** ****, *********** in ******* ** *** ground **** ***** **** would ***** ***** ******.

Reflections In Rain Cause False Alerts

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

*** ********** ****** **** had ****** ***** ****** on *********** ** ******* of ****.

Reflections-In-Nonmoving-Vehicles-Cause-False-Alerts

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

*******, ** *** *****, large ******* **** ******** as ****** ** ******** scenes. *** *******, *** dog ***** ******** *********, appearing ** * ******.

Large-Animals-Detected-As-Human

************, ***** ******* ********* triggered ******. *** *******, the **** ***** *** detected ** ***** ******** times ** *** ****.

Small-Animals-Detected-As-Human

Alerts ** ***** *******

*******, ***** ****** **** generated ** ***** ****** in *** ***** ******* changing, **** ** *** sun ****** ****** ******, causing *** ***** ** darken, *****.

Passing-Cloud-Trigger-False-Alert-Due-To-Light-Change

Vehicles ******** ** ****** / **** *****

****** *** ***** ** tested **********, ******** ********** of ****** ********** ** vehicles ** ******** ********** as ****** **** ****.

*** *******, *** ****** walking ******* *** ***** below ** ********** ** a *******. *** ******'* perceived ********** *** ** seen ** *** **** of *** ******** ***, much ***** **** *** person *******.

Human-Detected-As-Vehicle

*********, *** ***** ***** was ************* ** * person ** *** ***, though *** ****** ***** and **** ** ******* different **** * ******.

Vehicle-Detected-As-Human

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

****** *** *****, ** saw ** ********* ** DeepinMind ******* ****** ******* or ******* ******* *** scene, *** *** ** miss ******** ******* ******* the **** ** ********.

No-Missed-Person-Alerts-During-Day

*** *** *** *** miss ******* **** ******* in ***** ****, *****:

No-Missed-Person-Alerts-During-Night_Heavy-Rain

Demographic **** *******

******** ********/**** ******** ******* demographic **** ***** *** people ********, **** ** age, ******, *** ******* they *** ******* * backpack ** *******.

*******, **** *********** ** no ****** ********** ***** current ******** (**.*.** ***** 181028) *** **** ******** (v2.8.2.2).

Demographics Removed In New Versions

Version ****

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

  • ********* ***-*******-**/***: **.*.** ***** 181028

Comments (9)

Great review!  I hope you all will get a chance to test out their new bi spectrum thermal with those analytics.  I will be real curious how the thermal imaging analytics work in conjunction with the optical.

Great report.

Agree
Disagree
Informative
Unhelpful
Funny
Agree
Disagree
Informative
Unhelpful
Funny

Gary, thanks for bringing that to my attention, the link should be working now.

Agree
Disagree
Informative
Unhelpful
Funny

Previous firmware/iVMS versions offered demographic data about the people detected, such as age, gender, and whether they are wearing a backpack or glasses.

too bad, it might have been interesting to see the demographic data it came up with on the crow and the truck...

Agree
Disagree
Informative
Unhelpful
Funny

I'm yet to see a positive review about the DeepinMind range of products, which is a shame really when this is supposed to be Hikvision's answer to false alert minimisation. The tests above might as-well have been carried out on their entry range camera and NVR with the amount of false alerts triggered!

Agree
Disagree
Informative
Unhelpful
Funny

Ahmet, just to clarify, as we mentioned in the report, DeepinView camera did better than Deepinmind and DeepinMind still performed much better than the conventional camera /NVR analytics.

Agree
Disagree
Informative
Unhelpful
Funny

Sure, but when you look at a product like Netatmo for example, the ability to detect a human, car or animal or accurate every single time. It puts the likes of large 'CCTV' manufacturers to shame, hence my disappointment.

Agree
Disagree
Informative
Unhelpful
Funny

We found that to generally be the case - i.e., 'consumer' companies beating 'CCTV' manufacturers - Consumer IP Camera Analytics Shootout - Arlo, Google / Nest, Amazon / Ring, Hikvision / Ezviz

We only tested Netatmo once and that was 3 years ago, so I cannot speak for them. Generally, though, consumer cameras with cloud connectivity have the advantage of doing cloud false alarm filtering which has helped them (i.e., sent a potential alarm to the cloud first and run a check / deep learning to validate).

Agree
Disagree
Informative: 1
Unhelpful
Funny

Thanks for that, very informative.

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

This article is part of IPVM's 7,031 reports, 935 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