Video Analytics 101

By IPVM Team, Published Mar 16, 2020, 04:52pm EDT (Info+)

This guide teaches the fundamentals of video surveillance analytics.

IPVM Image

****** ** *****:

  • *** *** ***** *********
  • ***** ********* *******
  • ***** *** ********* *********? ******? ******? Cloud?
  • ******** ***-***** *********
  • *** ** *********? ********* *** **************
  • ****** ** ** *********
  • ***** ** *******
  • ***** ** ******
  • ***** *********
  • ** ****** ******** *********
  • ** ************** ******
  • **** ******** ** *** ******* ** Cameras
  • ******** *********** ** *******
  • *** ******** ********* ** ******* *** Recorders
  • **** ******** / ** ********** ** 2022
  • ****** ****** **. **** *******
  • ***** ******** ******** ********

Why *** ***** *********

************, ************ ****** ** * ****** to ******* ******** ******* ************** *** visually ******* **** *** ********* ** each ** ***** ** ****** ******. However, **** ** *********** *** ***** to ****** ******* ****** *** **** monitor ** **** ***** ** * time, ******* ** ****** ******.

***** ********* ******* ****** ******* ** detect ******** ******, **** ** ****** or ******** ******, ******* *****, ** license ******. ***** ****** *** **** be **** ** ******* ********* ** notify ** ******** ** **** *** effectively ******* **** ******* ** *** time.

Video ********* *******

****** ****** ** **** ******** *******, we **** ** **** *** ******* clear: ***** ********* *********** *****************, ******* **** **** ****** ** terrible. ***** ****** *** ***** ************ marketing ** *********, ** ** *** historically **** **** *************** ****** *** overstated ***********.

***** *** ****-********** ********* *********, **** more *** ******** *** ************ ** the **** *** ***** ***** **** further ***** ***********, *** ***** *** exceptions, *** *** ****.

Historic ******: *** (***** ****** *********)

***** ****** ********* (***) ** *** original ***** **** ** ******* ** try ** ****** ********. ** ** not ** *** '*****' *** **** may ****** ** ** ****.

*** ******* ** ********* ****** ********** changes ** ****** **** *** ***** to *** ****. ** ****** ****** change ** ****** ** * *****, the ****** ******** ****** ********.

*** ***** ***** ***** *** *** might '***' * *** ****** *** correctly ********* ******:

VMD ***** ** ***** ******

*******, *** ** *** *********** ****** to **** ******* ****** *** ***** movement ** ***, **** **** **** are ******. ******* ** ****, ** makes **** ******** *** ****** **** shadows, ******, ********, *******, *** ****** may ******* **. *** *******, *** image ***** ***** *** ****** * mistake ******* ** **** *** '****' whether *** ******** ****** *** * person ** * ******:

\

Basic ********* **** *******

** *** ** ****** ***** ******, manufacturers ********* ******* **** ********* ** remove ***** ****** ******, ******* ** basic ***** *********. ***** ******* *** include ****** **** ******/***** ******, ****** speeds, ********** ******, ***. **** ******** performance ******** ** ****** ***, *** false ****** ****/*** ***** **** ******, with *******, ******, ****, *******, *** other ****** ********** **** ********* (*** our****** ********* ********).

Why ** *********

******* ** *** *********** ** ******** VMD-based ********* *** *** **** ***** alerts *** ***, ** ********* **** developed *** ***** ************ ** ******** detection ******** *** ***** ************* **** classification ************.

*** **** *********** ********* ** ** analytics ** *** ****** ** ******** effectively ****** *** **** ******** ******: 'Is **** * ******?', '** **** a ***?', '** **** * *******?', 'Is **** * ***?' ***.

Basics ** ** *********

** ********* ******* **** ******** *********** by ******** '********' ** ***** **********. Simply ***, ***** ******* ** ***** objects *** ******** ** ***** **********, which ******* ***** ******** ********* ******* the ****** ** **** ** ** labeled **.

** **** * ****** ***** ** detect *****, **** **** ****** *** fed **** *** ** ****** ** it *** ******* *** ***** **** a **** ***** ****:

IPVM Image

**** ******** ** ********* *************, ******* an ******** ******** *** ** ***** aspects ** *******. *** ********* '******' on *** ***, ******** ** ****** ********.

AI ****** *********

*** **** ***** ***** ** ** is *********** ** *** ****** ** looking ** * **** ****** ** noise (****, ****, ******, *********). ***** more ******** **** ****** ***, ** is ***** ********** ** ************* ******* (rain, ****) *** ******** *******:

*** **** ***** ** ** ** determining ** *** ******** ****** ** a ****** ** ***. ********* ******** objects (******/*******/*******) ** **** *********** **** simple *********, *** ******** ******-***** *******.

**** **** *********** ** *********** ** the ****** ** * ******, *******, or ***** ****** ** ***.

IPVM Image

*******, ****** *** ******** *** *** most ****** ***** ***** ******* **** to **** ** ********* ** **** advanced *** ********* *******.

AI ****** **************

***** ********* *** ** ******* ** detect ******** ***** ** *******:

  • ****** - ***, ******, *********, ******** color
  • ******** - ****, **** (*** ** truck), *****, ********* ** ******
  • ******* - **** (*** ** ***), color
  • ********* ******* (****, ****) - ****, status (****-******), **** (****** ** *****)

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

******* * ******'* ******** ********** *** valuable *** ******* * ******** ****** during ** ***** ** ********, ** is ******* ** ******** ******, ***, clothing **** *** *****, *******, ****** hair, ***:

IPVM Image

*******, **** ** ***** *************** *** extremely ***********, *** ** **** ******* of ******** *****, **** ****** ************** analytics *** *** ****** ********.

Vehicle **************

* *******'* ******** ********** *** *** most ****** ******* ** ***** *********** about ** ********, ** **** ******* for * ******. ** *** ** trained ** ******** *** ******* ****, color, ****, ***** *** ********* ** travel:

IPVM Image

*******, ** ****, **** *** ***** are ********* ***********, ****, *** ********* marketed ** *********** ***/**** *************.

Animal **************

**** **** ****** **** ****** ** vehicle *********, ** *** ** ******* to ******** *** **** (***, ***), and *****.

IPVM Image

Inanimate ****** **************

********* ********** ******* *** **** ******** response **** ** *********** ********* **********. AI *** ** ******* ** ******** backpacks, *****, **********, ****, *** ******** specific *************** ** ****:

IPVM Image

*******, ** ****, ****** ********* *** classification *** **** *** ********* ******* by *********** *********, ***** *** ** expensive *** **** **** *********** **********. Classification ** ********* *** ***** ********* objects (**********, ********, ***.) *** ** very *********** ** ****, ***** ** lighting, ******, *** ********.

Color *********

***** *** ** ** ********* ****** for *********** ** ****** **** ********* for * ******** ****** ** *******. Color ********* *** ** ******* ** themselves ** ** *********** **** ***** objects (********, ******* *****):

IPVM Image

*******, ***** ********* *** ******* ************ in *** ***** ** *&* ** night *****, ********* ****** ** ****** of ***** *** ****:

IPVM Image

******

********* *** ** ******* ** ****** specific ***** ** *********. *** **** common ******** ********** ***:

  • ******** / **** ******** *** *********
  • *********
  • ****** ********
  • ****** **** ****** / *******

Tripwire / **** ******** *** *********

********* **** * ****** ****** ** approaches * ********** **** *** ** critical ** ****-******** *********, ** *** health *** ****** ********.

**** ** *** **** ***** ******** analytic *** ** ******* ** **** cameras:

********* ** **** *** *** **** reasons ** * ******** ********, *** the ***** ** ********* ** * person ******** ** ****, ****** **** just ******** * ****.

*********

******* * ****** ******** ** *********** an **** ** *** ****** ****** to ******* ** *****, ** **** are ***** ******** **** ******* ****** an ********** *****, ********* ****** *** offered.

******* ********* ********* ********** ** **** more *********** **** ********, ***** *** analytic ****** **** ***** ** ********* of *** ****** *** *** ******* time *****.

People ********

************* *** ****** ** ****** *** enter *** ***** ** **** ******* a **** *** ** ********* ** security ********** *** ****** ********, *** valuable *** ********* ** ****** *** hospitality *********.

****** ******** ********* **** ********* ** decrement ** ****** **** ******* * defined **** ** ****:

**** ** ******* ** ******** ********* but **** *** ******* ** ***** for ***** ****** ********.

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

****** **** ****** *** ***** ********* were ********* ******* ********** ******* (****, boxes, *********) **** *** **** ****** can ** ********* ** ***** ********* and ********* ************* **** ******** ** high-security ********* (********, *******, ********):

IPVM Image

*******, ******* ******-****-****** *** ***** ********* are **** *********** ** ******* **********, you **** ** ** ******* ** ensure **** *****, *** ****** ***** will **** **** *** ********.

Beware: **** ** ********* **********

********* ***** **** ******** ****** ***** also ****** *** *** ************** ** human ********' ******, ***, ******** *****, etc. *******, ** *** *******, **** the **** ********** ********* ***** *********** classify **** *********** ***********, **** *** classified ** *****, ******** ********** ** adults, ***.

IPVM Image

Where ********* *** *********

********* *** ** ********* ** *******, recorders, *******, ** *** *****.

*** ********* ** ***** * ********* are:

  • ******:******-***** ********* *** ** **** ******** because **** *** ** ********* ** low ***********, ****-******* *****, ****** ********. However, ******** ********** ******** *** ******** analytics **** ********* **** ******* ** expensive ******* ***/** *********** ************ (*.*. LPR/ANPR).
  • ********:********* ***** ** * ******** ********* allows *** * **** ******** ********* than * ****** ****** ***** *****, which ** **** *** *** ******* connected ** *** ********. ****, *** analytics *** *********** ********** **** *** viewing ********. *******, ******** ************* ** not **** ** ****-******* *** ******** to **** *** ********'* ***** ***, and ***** *** ********* *** *******.
  • ******:********* ********* ********** *** ***** ** have ********* ********* ********** **** *********, and ********* ******* ******-**** **********. *******, server-based ********* *** *********** ***** *** can ** ***** *** *****-*********.
  • *****: ********* ***** ** *** ***** should ********* *** ******** *********** ** recorders ** *******, *** ******* **** accuracy *********. *******, ***** **********, ************ public ***** ******* (******, ******), *** be **** *********. *****-***** ********* ******** typically **** ******* $**-$** *** ****** per *****. ********, ******** ****** *** will ******** *** *** ***** *** number ** ******* **** *** ** analyzed.

Deep ******** ** *** *******

*** ****** ************* ** **** **** learning ********* ** ***** ***** ************.

**** ********* *** * ***-******* ***** that **** *** ****** ***** ******** over **** ***** ** *** *****. However, * ***** ****** ** ******* and *******, **** ** ********'* ****-******** H5A ******, ******* ** ***** *** scene **** *** ****** ** **** a ****** ** ****.

Hardware *********** ** *******

*** ******* ********** ** ***** ** inside ******* *** **** *** **** of *** ******** ** *** ** inside *** ******. **** **** ** AI ******* **** ************ **** **** expensive. ******* ** ****, ******* **** VMD *******, ******* *** ************ *** realizing *** ******** ** **** ****** cost *** *** ******.

*******, ******** ****** **** ***** ********** and *********** ********** *************, ******* ** a **** ** ***-**** ****** ******** 'AI', **** ******* *******.

GPU ******** ********* ** ******* *** *********

******* ** ********* ********* ** ******* and ********* *** **** ******* **** camera-based ********* *** **** ** ******* multiple ******* **************, **** ******* * or **** *** ***** **** *** cost ********* ** ******* *** ****.

***** **** ***** **** **** ** analytics *** ********* *** ** *** reach ** **** ***********, ****** **** most ****** * *** ** ****-*** commercial, *********, ** ********** ********.

Deep ******** / ** **********

***** ***** ************ ********* *** **** promoted, *****, *** ******** *** ****** 20 *****,** ****, ** ******* **** **********,******* **** ****** *** ************* ***** fairly ******** *** ******** ****** ********* analytics. *** *** **** **** ** not ***** *** ** ********, *** other ***. **** ** *** ****. See ********** ********* ******** ******* **** ******* **** ** ****** manufacturers.

Single ****** **. **** ****** ***********

******* ********* *** ** ********* ** integrate ****** *** *******, **** ******* focus ** ******* ***** *** '*** to ***' ***** ********* / *** solutions. *** **** ****** ******* ** this ** ********, ***** ****** *** video ********* *** ******* ********** **** their******* ****** ***. ***** ******* ******** ******* *** analytic ****** ************* (*.*.,*******'* ******) ***** ********* **** **** **** their *** ***.

***** *************, **** ** *****, *********, and *******, **** ****** ********* ***** analytics ** ***** *** ********* (********* events, ******** *****, *** *************) ***** third-party ******* ** ******* ** * few ********* *** ****** ****.

Video ******** ******** ********

***** ** * **** ** ******* video ******** *********.

****** *************:

  • ********: ********* *** **** * **** focus ** *** ******* *** * number ** *****.
  • ***** *** *********: ***** ******** ** adding ** ** ************* ******* ***** low-cost *********.
  • **** *** ******: **** **** **** relatively **** ** ******* *********, *** are **** ******** ** ***** ********* to ******.
  • *****: **** ******* ***** ********** ********* for * ****** ** *****, *** typically ** ***** **** ********* ******. Recently **** ******* **** ******** ********* in *****-**** *******.

********-*****:

  • ******** - **** ****** ********** ******** but ** *********, ******* ** ****** systems (*** ****** *******).

********:

***** ********* ** *** ** ********* ************ **** *** *****, ****** ******** analytics ** ***-*** *** *****-****** *******. These *** ********* **** ****** ****** analytics **** ****** *************, *** **** often ***** *********** ********** (*.*. ****, weapons, ********, ***.).

Comments (6)

Great read, should generate some interesting comments.

Safety regulations are created usually after a bad experience creates the need.

Enhancements in analytics have happened after failures in the systems ability to meet the actual desired effect and the increase in techology hardware, a version of Moore’s Law.

The ability to over promise and under deliver is always present with analytics. Humans do that, not analytics.

I recall a product that continuously learned the environment (great idea for setup), had to add a feature ending that learning. Why? It was often used in static fence line environments and after a period of time it would forget what it was looking to detect.

Humans were “learned out.”

Anyway, it’s a lot better than it was and now we just need to create the “human filter.”

Agree: 1
Disagree
Informative: 1
Unhelpful
Funny: 1

Very good article. Funny how far ahead the hype is to the reality of what is shipping. I actually thought deep learning systems continued learning as time goes on as a standard to the term. Trip line is offered on many low end home type cameras such as Uniview and is pretty good in my experience. I think after reading this I would want a real time demo that I could inject possible false reporting situations to see how accurate the system would be. Thanks for posting.

Agree: 4
Disagree
Informative
Unhelpful
Funny

I can’t wait to see how far and how accurate Analytics will come in the next few years. Such a great tool once it’s a bit more reliable.

Agree: 3
Disagree
Informative
Unhelpful
Funny

A few notes:
1. Big difference between operation during day vs night. It may come as a shock, but people tend to try more intrusions at night/rain/fog. Night means not only no color and poor visibility - it means you depend on artificial lighting which is usually uneven, not optimally placed and creates reflections. Test your vendor at night time/rain.
2. "Left/taken objects" detection is harder than you would think. Very hard to differentiate between that and just 'some object was moved/lighting change".
3. Well made point about video quality - your camera equipment gets the high quality/raw video. Once compressed, many things become worse regarding detection. Raw is king.

Agree: 3
Disagree
Informative: 4
Unhelpful
Funny

In access control systems, occupancy reporting is generally tied to having an IN and an OUT badge reader at points of entry. The employees are required to actually badge out to leave the building. I like the idea of having the camera note who enters or leaves, the analytics can determine this based on direction of movement, and keep a running count of how many people are in the building. Nice :)

Agree
Disagree
Informative
Unhelpful
Funny

It's sounding a lot like what I always thought of as being video analytics, going beyond pixel change readings where you are trying to determine what the object is- person, car, etc, is considered now being called AI. And then from there it just goes to deeper and more comprehensive AI analysis, depending on resources and budget available.

Am I correct there...?

Agree
Disagree
Informative
Unhelpful
Funny
Login to read this IPVM report.
Why do I need to log in?
IPVM conducts reporting, tutorials and software funded by subscriber's payments enabling us to offer the most independent, accurate and in-depth information.
Loading Related Reports