Testing Color Analytics Performance

By: Ethan Ace, Published on Dec 16, 2013

In an industry known for overhyping its own value, color analytics have been considered pure science fiction, with few, if any, stories from the field to back up manufacturer claims.

In this report, we test Bosch's IVA color analytics to see which practical applications they performed in, and where they failed.

First, in an interior conference room with a human subject:

And an outdoor parking lot/driveway with different colored vehicles:

We varied the precision of the analytics to understand how performance varied, from least to most sensitive using their built-in scale:

**** ************* *** ********** *** own *****, ***** ********* have **** ********** **** science *******, **** ***, if ***, ******* **** the ***** ** **** up ************ ******.

** **** ******, ** test*****'* *** ***** *********** *** ***** ********* applications **** ********* **, and ***** **** ******.

*****, ** ** ******** conference **** **** * human *******:

*** ** ******* ******* lot/driveway **** ********* ******* vehicles:

** ****** *** ********* of *** ********* ** understand *** *********** ******, from ***** ** **** sensitive ***** ***** *****-** scale:

** **** ****** * variety ** ****** **** red, *****, ****, ******, neons, ***.

[***************]

Key ********

  • ** ****** *** *****, cameras *** ****** ** lighter **** *** ***** eye, ****** ******** ********* of ****** *** **** difficult. **** *******, ******* color, **** **** ** simply ****.
  • ******* ********* *** *** results ** ****** ******** to *** ******* *** on *** ***** ***** triggering ****** (*.*., ****** and **** ********** **** blue ** *******).
  • ******************* ** * (********), only *** **** ***** minimum, ******* *** ******** of ***** ******* **** other ******, *** **** increases ****** ** ****, as ****** **** ** very ******** ** ***** to *******.
  • ******************* ** * ** 4 (**** ** **** high) ******** ** ******* numbers ** ****** ******* (75% *** **) ** alerts *** ********* **** on * ***** ***** of ****.
  • ** *** *****, ********* over ****** ******** ******* in * ****** ***% failure ** ******, ** the ****** ****** ******* colors ********, ****** *** human *** ***.
  • ** * ****** ******* parking *** *****, ******** were ************ ******** **** using ********* *. ** precision * *** *****, rules ****** ** ******** at ***. 

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

***** *** ********* ******* between ******** *** ***** alerts, ** ** *** recommend ***** ********* ** used ** ******** ************. Alerting ** **** ******** colors **** ****** ** many ****** ******* *** to ****** *********** ** shade, ***, ** ******** lighting *** *******. *******, lowering ********* ******* ** numerous ****** *** ****** other **** **** *** be ********. ******* ** this, ***** ********* ****** only ** **** **** looking ** ****** *** number ** ****** **** in **** ******** ************.

It ***** ** *****

*** ***** ********* **** were ***** ** ** a **** *** ****** (the ***** ***-****) **** *** quite **** ** *** ***** ******** **** *** *** ****** *** ***** ***********. *********** ***** ****** be **** ***** ** lower ******* ******* *** used **** ****** ***** color ********* **** **** to **** **** ******* cameras *** ****** ****** differently.

Configuring ******

**** ********** ******* ************* of ***** ********* ** well ** **** ****** issues ** *** **** color ********:

Selecting ******

*** ******* ******** ** configuring *****'* ***** ********* properly ** ************ *** the ********** ******* **** the ****** "****" *** the ***** ******** *** the ***** *****. *** example, ** **** *****, the ****** ******* *** in *** **** *******, as **** ** *** object ********** ** *** lower *****, *** ******** the ********* ****:

*******, ** ***** *****, the ****** *** ** red, ****** ****** **** instead, ******* ** ******:

** ***** *** ******* that **** *** *****:

  • *** ****** ** ****** the *******'* ***** ** the ******** *****, ***** it ****** ******.
  • ** ***** *****, ********** and ***** ***** *** bleed **** *** ****** properties ** ** ****** across *** **** ******.

*******, **** ******* *** bright, ******* ********** ** the ***** ** *** conference ****, *** ****** often ******** **** ****** as ****** ****, ********** of ****** ***. **** was **** ** *** only ***, *** ****, green, ******, ******, ***.

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

** *** *****, ********* had *** **** ********* with **** ****** **** as *****, ****, *** brown. ****** **** *** simply ******* *** ****** never *** *** ***** as *** **** ******** seen ** *** ***** wheel, ****** *** ** variances ** ********** **** lights, ***.

*** *** **** ****** are ******** ** *** color ***** **** *** very ******** ** ******, as **** *****. ****** the ******** ****** ***** while ********* ***** ** the ****** ** *** wheel ******* ** **** or *****. *******, ** also *********** ******* *** colors ******* ** **** those **** *** ****** of *** *****, ****** triggering ********** **** **** difficult.

Color ********* **********

*** ***** ********* ******* allows ******** **** ** be ******** **** ** just * *** ******, intended ** ******** ********. This ******* ****** **** 0 (*** ***** *******) to * (**** *** specific *** *******), ** seen *****:

** *** *****, ** found **** ********* ***** 1 ******* ******* ******** rates. *******, ** **** level, ****** ***** **** the ******** *** *** trigger ******, ** ****** and ****** *** ******* when *** ** ********, for *******. ** ********* 2 *** *****, *******, matching *** ************, **** below *** ***** **** with ********* *, ** the **** ********* ** all, ** **** ** examples *****.

Color ********

** ***** ** * well *** ****, ***** 160 ***. *** ******* walked ****** *** **** ten *****, ********* ** the ********* *******. ***** rates **** ************ ********** regardless ** *****, ********** at ********* *, ***** triggered ** ***% ** attempts.

*****, ***, *** **** all *****************, **** ***% ******** at ********* *, ******** to **% ** ********* 2. ****** ****, ** precisions * *** *, exact ****** ****** *** specific ** ******* *** rule:

****** **** ****** **** as *** ****** *** green ******** ***** **** performed ** ***% **** precision *, *** *** not ******* ** *** at ********* * ** above. **** *** ******* the ****** ***** *** these ****** ** ****** grey ** *****, ***** are ******** ** ****** precision ******.

Low *****

******** ***** ** ***** 2 ***, ******** ** even ********* * ***** substantially ** ****** *** cases. ** **** ***** level, ** **** ***** adjusting *** ********** ****** to ********** *** ****** appearing **** ** *** light, *** **** *** no ****** ** *********** whatsoever.

*****, ***** ***** ** about **% ******** ** precision *, *** ** undetected ** * *** above:

 

********* ** *** ******* most, **** ***** ***** at ********* * ******** to **** **%, **** no ******** ** * or *****:

**** ********* ****** **** either *** ** *****, maintaining ***** **% ********, though ***** ******* ** match ****** ********* *:

**** ***** ********* **** out ** *** ****** in *** *****, *** to *** ******* **********. At ********* *, ** remained ***% ********. ********** to ********* *, ** maintained * **% ***** rate, ****** **** *** other ***** ******:

*******, ****** *** ** matching **** ****** **% at ********* * ** dim *****, *** *** not ******** ** * or *****: 

Vehicle ********* *********** 

** ********* *, ******** were ************ ******** *** to ***** ****** ****. However, *** ** *** vehicle's ********, *** ***** changes ** ********* ******* are ********* *** ******* move, ****** *** **** narrow ***** ****** ** precision * *** ***** fail ** *******. **** was **** ***** ******** colors ** *******, ********* blue:

*** ***:

Comments (19)

Can this form of analytics be more usefull with thermal units delivering a colorized output?

Hello Birger.

No, unless you use radiometric thermal cameras giving the absolute temperature. The "standard" thermal cameras for video surveillance are differential, not radiometric: this means that they percept and show bodies with different temperatures, but you don't know if that temperature is 37° or 15°.. You just know that it is different, and the false colors (never use them with analytics; always black/white mode) are definitely relative and built frame by frame by the camera normalizing the whole image according to the different temperatures of the different bodies present in the scene in that moment.

This means that those colors are even much less reliable to search and track than when you search a color in an optical camera.

Cheers,

Simone

(TechnoAware)

Thank you for a very informative answer Simone.

/Birger

Finally!

Thank you so much John for this great report!

I will forward it to all people against whom I "fight" daily trying to convince them how much it is difficult to consider "color" as a reliable feature for analytics! And to all the writers of tenders who believes always to unlikely commercial brochures (not only from Bosh of course!..) without simply reasoning on the practical phisical issues..

Have a great time,

Simone

(TechnoAware)

Thank You Ethan & John

Great Report , Goes beyond the traditional reviews of the color dynamics of past reviews .

Great Job

Curious how it would work in a "perfect" scenario?

ie: assembly line, parking lot entrance/exit, doorway, etc. Basically if you had a setting with perfect lighting and small FOV.

Chris, parking lot entrance/exit and doorways are going to be especially challenging as varying light levels and strong backlighting are going to change the color appearance of objects over time.

I was thinking of possible applications that were made "perfect" vs "real world" applications.

Where you could control lighting and the "scene" at all times, possibly even the colors (no variations).

Yes, so if you can control things 'perfectly' it should work well. For example (I am making this up but I suspect this would work well), you have a conveyor carrying fruit and you want to automatically detect and flag fruit with 'off' colors. In such a scenario, you could carefully set it up and optimize it to accurately identify those off color fruits.

Yes, that would be a fine example.

So in a "perfect" scenario it should have a high % of accuracy otherwise don't bother...

I would assume it would also be best to pilot test the application before selling it ;-)

Chris, actually the point it's exactely this one you wrote, about making tests before.. Actually it's the main problem of this analytics market.. The problem is when the "smart" commercial always makes tests or shows demos in perfect scenarioes, with perfect stable conditions, for few minutes.... But unfortunately maybe that is not the reality.. Maybe that's just a trick, because unfortunately in the real world then the conditions are slightly never that stable for long time..

But after buying, the Customers verify in real life that it's all fake, that instead nothing works, and then none wants to see analytics anymore....... Of course Customers watch CSI in the evening and they want miracles; so "smart" commercials of course feel in need to show miracles.. But maybe it's not that "smart" and healthy to go on that way..

There are for example still many many products working by fixed background reference.. And believe me, there is no better method at all if you have 5 minutes of indoor perfectly stable and controlled environment, no shadows, no reflections, perfect stable light.. Uh, you can count the hair of a person by that method!.... But for 5 minutes... ...maybe! Nice trick..

As I always say to our Partners who tell to me "oh but what you say, I saw it by my eyes!....": yes, me too yesterday evening I was in a show of a good magician and I saw by my eyes a woman divided in 2...... But that doesn't mean the woman was really divided in 2, right?....;)

Cheers,

Simone

You got it ;-)

Guess that's why we are all here (reading IPVM), we appreciate the work they area doing in clearing the smoke screens and verifying reality...

We are using the color filter mostly for searching in the recorded video. It can help a lot to retrieve a certain person.

Thank you Bosch employee for that vague undisclosed endorsement. Care to elaborate on what people, situations, accuracy, and time frames you are doing this in?

John, perhaps you should add the additional caveat of 'Identity will be known to site admins' right after the "Post Without Disclosing Your Name" checkbox.

Some people must think they are truly anonymous, how else can you explain their audacity?

How's that? Guess we have to buy Micro SD cards then and record on the camera then run the analytics after the fact?

How does that help with alerting of an issue while it's happening vs hrs/days/weeks/etc after when it's realized, if the video is still on the card?

I would seem to think that analytics would be most powerful being run pro-actively vs re-actively, no?

Good luck finding which box the off color fruit is in, after you finally get time to run the analytics against the recorded video...

Yes, for "forensic analysis"...... Of course, because the issues of color when they come from an off-line video instead of a live one magically are solved, right?.. Wow, magic hard disk!....

Or maybe he was meaning that they open the off-line video in a Video Editor, they go to paint with bright red color the dresses of the person they want to search, then they run the analytics searching for red color and say "Oh! That's it!!..."........;)

The typical rejoinder for search is that it does not need to be as accurate as alerts. For example, if I search for a red car and get 50 matches and only one is really red, I can quickly scan through the 50 thumbnails rather than scan the full video over hours. This of course is premised on the fact that the true red car is not missed entirely.

But that's exactely the point John. As you correctly showed, the main problem is to see the color and find it.. Even in indoor as you showed, and we see then outdoor what it happens: when there are shadows, camera over/under-exposures, same camera light compensations and automatic gains, reflections of sun or any light, with artificial lights, by night, ...

I absolutely don't say it's useless, not at all. There are many examples of valuable possible applications, given a priori timely simplifying assumptions in the environment or in the application itself.. Infact also us we have this color filtering function. Only, as usual in the market there's a lot of science fiction and this kind of reports help to heal this market from confusion and false expectations..

Thanks, cheers

Simone

(TechnoAware)

Login to read this IPVM report.
Why do I need to log in?
IPVM conducts unique testing and research funded by member's payments enabling us to offer the most independent, accurate and in-depth information.

Related Reports

Panasonic 4K Camera Tested (WV-S2570L) on Sep 13, 2019
Panasonic has released their latest generation 4K dome, the WV-S2570L, claiming "Extreme image quality allows evidence to be captured even under...
WDR Camera Mega Shootout 2019 - Avigilon, Axis, Bosch, Dahua, Hanwha, Hikvision, Panasonic, Vivotek on Sep 04, 2019
This is the biggest WDR Shootout ever, with 23 cameras from 8 manufacturers: We included true multi-exposure WDR and non-WDR models, ranging...
AI Video Surveillance (Finally) Goes Mainstream In 2020 on Sep 03, 2019
While video surveillance analytics has been promoted, hyped and lamented for nearly 20 years, next year, 2020, will be the year that it finally...
Scylla AI Video Analytics Company Profile on Aug 29, 2019
Scylla, an AI analytics startup, says they are targeting 1 Billion dollar valuation in 5 years and it "is not rocket science" to detect weapons and...
Uniview 4K Camera Tested on Aug 29, 2019
In our 8MP / 4K Fixed Lens Camera Shootout, Uniview's low end 8MP was firmly in the middle of the pack. Does their higher end Pro Series 4K camera...
Wyze Image Quality Tested vs Axis, Hikvision And Nest on Aug 27, 2019
With Wyze's new AI feature, it is strong at analytics, but how good is the camera's image quality? Is it just as good as incumbents like Axis,...
Anyvision Facial Recognition Tested on Aug 21, 2019
Anyvision is aiming for $1 billion in revenue by 2022, backed by $74 million in funding. But does their performance live up to the hype they have...
Dahua 4K Camera Shootout on Aug 20, 2019
Dahua's new Pro Series 4K N85CL5Z claims to "deliver superior images in all lighting and environmental conditions", but how does this compare to...
Verkada People And Face Analytics Tested on Aug 16, 2019
This week, Verkada released "People Analytics", including face analytics that they describe is a "game-changing feature" that "pushes the...
Axis 4K Camera Shootout 2019 on Aug 14, 2019
Axis' 4K Q3518-LVE claims the "best video quality possible", with Lightfinder super low light performance, Axis' high end Forensic WDR, and...

Most Recent Industry Reports

How Cobalt Robotics May Disrupt Security on Sep 13, 2019
While security robots have largely become a joke over the last few years, one organization, Cobalt Robotics, has raised $50+ million from top US...
Panasonic 4K Camera Tested (WV-S2570L) on Sep 13, 2019
Panasonic has released their latest generation 4K dome, the WV-S2570L, claiming "Extreme image quality allows evidence to be captured even under...
ASIS GSX 2019 Show Report Final on Sep 12, 2019
IPVM went to Chicago for ASIS GSX 2019, with many exhibitors disappointed about traffic and the exhibitor schedule changing next year. Inside we...
Installation Course - Last Chance - Register Now on Sep 12, 2019
Last Chance - Register Now - September 2019 Video Surveillance Install Course. Thursday, September 12th is your last chance to register for the...
Commend ID5 Intercom Tested on Sep 12, 2019
Commend touts the new ID5 intercom as 'timelessly elegant' and the slim body, glass front touchscreen indeed looks better than common, but ugly,...
US State Department: "Chinese Tech Giants" "Tools of the Chinese Communist Party" on Sep 12, 2019
The US State Department has called out "Chinese tech giants" for being "tools of the Chinese Communist Party" in a blunt new speech that makes...
Uniview OEM Directory on Sep 11, 2019
This directory lists 20+ companies that OEM products from Uniview, with a graphic and links to company websites below. It does not cover all...
Yi Home Camera 3 AI Analytics Tested on Sep 10, 2019
Yi Technology is claiming "new AI features" in its $50 Home Camera 3 "eliminates 'false positives' caused by flying insects, small pets, or light...
Hanwha Announces 32MP Camera + AI Line on Sep 10, 2019
In the first rise in maximum megapixel resolution in 5 years, Hanwha has announced a 32MP / 8K camera directly competing with Avigilon's H4 30MP /...
Fingerprints for Access Control Guide on Sep 09, 2019
Users can lose badges, but they never misplace a finger, right? The most common biometric used in access are fingerprints, and it has become one...

The world's leading video surveillance information source, IPVM provides the best reporting, testing and training for 10,000+ members globally. Dedicated to independent and objective information, we uniquely refuse any and all advertisements, sponsorship and consulting from manufacturers.

About | FAQ | Contact