Vivotek "Neural Network-Powered Detection Engine" Analytics Tested

By Rob Kilpatrick, Published Sep 17, 2019, 11:40am EDT

Vivotek has released "a neural network-powered detection engine", named Smart Motion Detection, claiming that "swaying vegetation, vehicles passing by, or animal activities in the scene will no longer trigger" events.

free image2

IPVM tested this on Vivotek's FD9391-EHTV, examining:

  • How accurately does it detect humans walking, running, and crawling through the scene?
  • Does it avoid false alerts on animals, foliage, rain, and other common outdoor sources?
  • Does it avoid false alerts on shadows, lights, and other common indoor sources?
  • How far can it detect people?
  • How difficult is Smart Motion Detection to set up?
  • How well does it integrate with various VMSes?

*******

*******, ***** ****** ********* was **** ********, ********* people ******* *** *******, day *** *****, *** avoiding *** **** ****** indoor *** ******* ***** alert *******, **** ** shadows, *******, *******, ******, and ***** ****/******** ** the ****. *** ** these ******* ******** ********* many ***-** ********* *** even **** **-***** ****.

*******, ** ****** ** detect ****** ******** ******* the *****, ***** ******* says ** ****** * training ***** *** *** be ********* ** ****** versions.

************, **** ********** ** dwell **** *** *********** were ******** ** ********* false ****** ** *** test ******, ****** ***** adjustments **** **** ****** time *** *** *** require *** ******* *********** or ********.

*******, ***** ****** ********* integrated **** ********* ********, Genetec ******** ****** *** Vivotek's *** ***** ***, but *** *** ********* in ********, *****, ** Nx *******.

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

***** ****** ********* ** available ** ** ****** on ****** ******* ******. It ** ********* ********* in * ******, *** Vivotek ***** ** *** the ******* ** *** models ** **** *** released.

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

****** *******, ****** ******* upright ******* *** ***** were *** ****** *** or ***** **** **** walked ******* *** ******** calibration ****.

Vivotek-Smart-Motion-Accurately-Detects-Humans-Walking

****** ******* ******* *** scene **** **** ******** accurately.

Running-Subjects-Detected-Accurately

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

*******, ***** ****** ********* does *** ********* ******** subjects. ******* ***** **** this ** * ***** limitation ** *** ******* version *** ***** ******* people ********* ****** ******* is ******* ** ****** upright ******* ******* ****. [[check ****: **** **** to ********* ******** ********* in * ****** *******.]]

Crawling-Humans-Missed-Completely

Most ******* ******* ** ***** ****** *******

********, ****** ******* ** false ****** **** ******* by *******'* ***** ****** Detection, **** ** ***** animals ****** ******* *** scene (* ***** *** shown *****).

No-False-Alerts-From-Large-Animals

*** ******** **** *** not ******* ** ***** animals, **** ** *** cat ***** ****:

No-False-Alerts-From-Small-Animals

*******, ******* *******/******** **** not ******** ** ******, nor **** **** ** an ******** ****.

No-False-Alerts-On-Vehicles-Or-Foliage

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

***** **** *** **** drops ** *** **** did *** ******* ***** person **********, *** ** the **** ****** ***** alerts *** *********.

No-False-Alerts-On-Heavy-Rain

Common ****** ******* ** ***** ****** *******

*******'* ***** ****** ********* ignored *** ****** ******* of ***** ****** ******* such ** ******* ***** changes, ***** *****, ***** frequently ********* ***-** *********.

No-False-Alerts-From-Massive-Light-Change

************, ******* *** ****** going **/*** ** ******** rooms **** *** ********, also ****** ** ***** analytics.

No-False-Alerts-From-Shadows-And-Lights

False ****** ** ****** ****/********** ****** ******** ************

*** ** *** ******* above ******** ** ******** changes ** ******. *******, in *** ******* **** the ****** *** ***** near * **** ******, false ****** ********** ************ occurred. ******* *********** * slight ******** ** *********** and ** ******** ** dwell ****, ***** *********** stopped ***** **********.

Person ********* ** ~***'/~** ***

** *** *****, ***** Motion ********* ******** ******** people ** ~***'/~** *** on *** ** ******-****. However, ** ** ****** that ******** ** ********* on * ***** ********** stream, ****** ****** *** requirements *****, *** ** could *** ****** ****.

Humans Reliably Detected At 125'  (25 PPF Using 4K FD9391-EHTV)

Simple *****

***** ****** ********* **** requires *** **** ** input *** ****** ** the ******, *** * region ** ********, *** toggle ****** ********* **. The ****** ************* ******* camera **** *** **** using * ***** ** gyro ****** *** ************* calibrates *** ****** ********* area (*** **** ***** below).

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

*** **** ****** ***** which ********* ***** ****** Detection **** ********* ******** and ******* ******** ******. It *** *** **** in ********, *****, ** Nx *******.

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

***** ****** ********* ********** with ******* ******** ****** using ****** ****** ******, the **** **** ****** unit **** ****** ********* and ****** "****** **" and "****** ***" ** event *******, ****** *** then ***** ** *** event **** **** * person ** ********.

Integration **** ******* *****

***** ****** ********* ***** integrates **** ******** *** VAST2 ***, ********* ***** search *** ****** **********, viewable ******** *****/****** ******, and ****** **********.

Integrated-With-Vivotek-VAST2-VMS

****: **** ********* ******* VAST2 *** ****** ********** the ****** **** ******* a ****** ******* ********** can ********* ***** **** of *** *******'* ****. This ****** ********* ** turned ***.

VAST2 Dot Circle Obscures Object (Cannot Be Disabled)

Teardown - ********* *** ****

*** ******* ******-**** **** a********* ****** **** ****** SoC***** ******** * ****-**** ARM ********* *** ************ *********** ** ********** ******* intelligent ******** ************ **** as ****** *********, ********* defense, *** ***** *********

Versions ****

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

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

Comments (11)

Agree
Disagree
Informative
Unhelpful
Funny

So failure to detect crawling people is a training issue? I guess that makes sense. The neural networks need to be trained in what to detect and they weren't trained for crawling people yet.

But that brings up some big questions...

If people don't walk in one of the ways the system has been trained for, will they go undetected?

In other news, MI6 announced a new program for thwarting analytics:

Agree: 1
Disagree
Informative
Unhelpful
Funny: 10

Hi Undisclosed Integrator #1,

VIVOTEK's Smart Motion Detection actually detects the "shape" of human and it is able to detect so as long as the upper half body is shown in the detection area. We plan to secretly train our staff the Silly Walks and see how well our AI engine works.

Agree
Disagree
Informative
Unhelpful
Funny: 4

Update -

Added section showing the SoC Used: The Vivotek FD9391-EHTV uses a Hisilicon Hi3519 V101 camera SoC which includes a dual-core ARM processor and an integrated FPU claiming to supporting various intelligent analysis applications such as motion detection, perimeter defense, and video diagnosis

Agree
Disagree
Informative
Unhelpful
Funny

VIVOTEK will release multiple NDAA compliant models that support Smart Motion Detection. Please see more details at Vivotek NDAA Statement.

Agree: 2
Disagree
Informative
Unhelpful
Funny

How did you get such a square box arround the human?

Is this something in camera or is this only compatible in the Vivotek VAST2 VMS?

Agree
Disagree
Informative: 1
Unhelpful
Funny

How did you get such a square box arround the human?

Is this something in camera or is this only compatible in the Vivotek VAST2 VMS?

It's the way Smart Motion Detection shows bounding boxes when a human is detected, only viewable in the web interface or in VAST2 live view/playback.

Agree: 1
Disagree
Informative: 2
Unhelpful
Funny

Thank you for your reply!

Did you test the human detection at night with the built in IR also. How far could the camera detect at night?

Agree
Disagree
Informative: 1
Unhelpful
Funny

Did you test the human detection at night with the built in IR also. How far could the camera detect at night?

Yes, it was tested at night, detection was the same night/day, the camera needed ~25 PPF to detect a person in both scenes, though a night scene might need a slight motion/person sensitivity increase to keep tracking at the threshold.

Generally if we see a decrease in performance at night, it will be noted in the report.

Agree: 2
Disagree
Informative: 2
Unhelpful
Funny

This is interesting. I can use this for fenced perimeter because crawling wouldn't really be an issue from the inside of the campus. Not a bad price.

Agree: 2
Disagree
Informative: 1
Unhelpful
Funny

Update:

Updated VMS Integration section to add integration with Genetec Security Center. Smart Motion Detection integrates with Genetec Security Center using normal motion alerts, the user must toggle unit side motion detection and enable "Motion on" and "Motion off" in event options, events are then shown in the event list when a person is detected.

Agree: 1
Disagree
Informative: 1
Unhelpful
Funny
Read this IPVM report for free.

This article is part of IPVM's 7,334 reports and 972 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