Detection, Classsification, Recognition and Identification (DCRI)

Author: John Honovich, Published on Feb 04, 2013

Detection, classification, recognition and identification is all very easy until you put it in practice. Then you realize that there are big practical and subjective problems that make using them difficult.

Academically, the terms are straightforward and reflect what they mean in real English:

  • Detection - The ability to detect if there is some 'thing' vs nothing.
  • Recognition - The ability to recognize what type of thing it is (person, animal, car, etc.)
  • Identification - The ability to identify a specific individual from other people

Obviously, each level is harder and requires more detail, going from detection to recognition to identification.

*********, **************, *********** *** ************** ** *** **** **** ***** you *** ** ** ********. **** *** ******* **** ***** are *** ********* *** ********** ******** **** **** ***** **** difficult.

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

  • *********- *** ******* ** ****** ** ***** ** **** '*****' vs *******.
  • ***********- *** ******* ** ********* **** **** ** ***** ** is (******, ******, ***, ***.)
  • **************- *** ******* ** ******** * ******** ********** **** ***** people

*********, **** ***** ** ****** *** ******** **** ******, ***** from ********* ** *********** ** **************.

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

**** ************* ******** ***** ********** ********* ** **** ******** *** * ***** ***** *****. *** ********, ****'* *********** **** ***** **** ** ******* ******** *******:

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

**** ****** * ******** *** ****** ** * ******* *****. While ***** **** ******** *** **** **** ****** *****, *** two ************* ******* *** *** ***.

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

  • ***** ********** ~*****
  • **** ********** ~*****
  • *** ******* ******** *** ********** ~****** (***** ****** *** **** paper)

**** ***** ******** ***** **, "**** ** ***** ** ****? Is ******* *****, ***** ** ********?"

*** ** **** ***** ** *** **** ************, ***, *************, two *********** ****** ***** **** ***** ***** *******.

2 *********** ******

*****, *** ***** *** ********** ********** *** ****** ** *** perception *** *********** ** *** ********** ******.

**** ** **** ****** *** '**************' *** **** ************ ***** people. ** *** ********* *** *************, ** **** **** ********** debates ***** *** ***** **** ****** *** **************.

*** **** *** ********. **** ** *** * ****** **** by **** *****:

***** ***,*** ***, ** ********** *** **************? ** (** *** ****), ** (in *** ******) ** ** (** *** *****)?

** ** **** *****, ** *** * ***** *** ***.

*** **** *** **** ********** ** *********** ****'* **** ****** for *********. ***** **** ***** ***** **** ** ****:

**** ***** **** ************ *********** ** **** ********* ****** ******* in ****** ******* ******** *** **** *** ********** ** ** the ****** ** ****** *** ****, ***** *** **** * massive ****** ** ****** ******* *** ******* ******.

*******

**** ****** **** ** ** ***** ** **********, *** ****** often ******* '*******'. ** ** **** **** **** ****** **** other ****** **** ** ******* * ****** ** **** ******* many ****** **** ********* ***********.

*** ** *** ******* **** **************? ** ** * ***** call *** ********** *** '******' ** ******* ** ****** *** the ******.

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

*** ***** ***** ** **** *** **** ****** ** ****** will ******* ********* ***** ******* ********* ** *** **** ** day. *** ***** ** ** ** ******* **** ******** ********** with ***** *** ************** *** ** *** *** **** ** rises ** **** **** *** ****** *** ***** **** ****. Worse, ** *** **** **** ****** ******** ** *****, *** still ***** **** ***** ** **** ****** *** ****.

What ** **?

***** ***** ***** *** ******** ***** **** ** * **** is ******. **** ** *** ***** **** ** ****** **** that ** ********** ** ****** ****.

**** ***********, *** *** ***** ************:

  • **** ***** ** ******* **** *** ****/*****/******** ***** ****** ****? Decide **** * ****** *******, ***'* **** *** '*********' *** hope **** ***** **** *** ****** ** ************.
  • **** ******** ** ************* ****** *** ***** *** *** **** of ** ****** **** ** ****? **** **** ***** ** front **** *** ****/*****/******** ****** ** **** ********** *** ******** challenges ** * **** ***** *********.

Comments (16)

Is there a ppf threshold that facial recognition analytics need to be effective? i.e. how many ppf are required before analytics can differentiate one face in its database from another?

Good question. Typically, ppf requirements for face recognition analytics are significantly higher than identification.

Face recognition systems typically use a slightly different pixels metrics. Instead of ppf, they generally cite pixels between the eyes. However, that can be easily converted to ppf. Assuming 2.5 inches between the eyes, multiply the pixel between the eyes metric by ~5. A common facial recognition metric is 35 pixels between the eyes, which means 175ppf.

However, some systems ask for even more. We've seen requirements for 50 pixels between the eyes (e.g., 300ppf, requiring a 1.3MP camera to cover a tiny 4 foot wide FoV).

Indeed, one vendor recently said they can do it with just 22 pixels between the eye, which would be 110ppf, lower than what the Sweden National Lab claims for a human to identify a face.

All that said, computers have much lower tolerance for variation and fuzzy images so expect to provide a lot higher PPF for facial recognition analytics.

John,

Of course, many specifications still refer to the Johnson Criteria from before you were born. That creates it's own set of hardship. When they say detection, that can be as little as 1.5 pixels. Recognition can be as litttle as 6 pixels. Recognition can be 12 pixels. That criteria is based on seeing anything at all (Detection), to knowing it's at human, vehicle etc. 2 pixels....LOL

Greg

Greg, thanks, here's a 2 page overview from FLIR on it :)

I had been looking for a more detailed technical paper but could not find one yet on the Internet that explained it detail how it was designed / determined. Anyone who can find, please share.

The problem I see with the Johnson criteria is the same fundamental one as for any conventional surveillance camera, specifically the subjectivity in saying how much is enough to 'detect' a person.

For instance, here's Vumii's video of detecting at 1.5 pixels.

Maybe my eyesight is gone, but that is incredibly difficult to tell that there is an object there. Even if they could spot someone, the level of eyestrain and difficulty must be severe for anyone trying to do this on an ongoing basis. I have to imagine many people would look at that and say, "I need more details / more pixels on target"

Thanks for this, John. I had to look very hard at this and I have years of experience forensically watching very low quality video, a ton of which was on VCR tapes. Every time the tape was played the video was degraded. These people in the above video could easily be mistaken for tape "wow and flutter" if on video tape and could be mistaken for low resolution shifting seen on compressed video archives. I find it extremely difficult to believe that a live person monitoring this video would detect that movement unless they were ONLY monitoring the one camera, AND the image was full screen!

John,

The Johnson Criteria should have been updated many times over. The function of DRI has a completely different meaning today. Imagine the technology of CCTV in 1958. 330TVL was amazing!

It's time to update the criteria. I really don't know anyone who argues to retain it, but it is the only imperical reference for US companies. I hope you don't think I'm defending it.

The concept of megapixel imagers for security wasn't even a pipe dream in that time period.

Greg

Let's replace the Johnson Cirteria with the Johonovich Criteria!

: )

Regarding face rec criteria, there is also a differnce between having enough pixels to to do a face search that brings back nearby similar faces, vs, face authentication, which claims with a high degree of confidence that a face is a particular person.

Gator

The UK has used a human size/aspect ratio test chart called a Rotakin for testing end-to-end cctv performance for many years. It is a practical method of testing for identificaton criteria. It was developed by the UK Police Scientific Development Branch

A Rotakin testing requirement was written into the UK minimum survellance standards for all licensed CCTV installations although I am not sure if this is still used.

Re: Rotakin, we had a recent discussion on it. None of us could understand the value of it, save for in the UK, it has a legal basis. Otherwise, it's just another chart that is still subject to issues of lighting variations and differences in perception about what is good enough for detection, identification, etc.

Hello all!
Let me see if I can shed some light on this.

Identify, recognize, detect, etc. come from European standards published in 1996, see http://www.cenelec.eu/dyn/www/f?p=WEB:110:956246286239172::::FSP_PROJECT, FSP_LANG_ID: 5028.25 (do not buy this because it recently (Jan 2012) has been replaced).

From there you can read that the possibility of such identify a person for an operator (i.e. primarily for live viewing) is that the person occupies 120% of the image height (cut off at the knees). And to be able to detect a person is thus about an operator, at the right distance from the monitor, reliably shall be able to detect a person moving in the picture. The prerequisite is that the image resolution is 400 TV lines or more (scanned in the horizontal plane). After these basic requirements, then also lenses, lighting, etc. has to be taken into account. This works quite well.

Regarding Swedish National Lab recommendations then these are actually from - the FBI! Note that these recommendations are really just for commercial establishments (ie indoors). And this gets very messy because it does not comply with European standards (are you still with me?).

By the way, if anyone wants a copy of the FBI's "Recommendations and Guidelines for Using Closed-Circuit Television Security Systems in Commercial Institutions" just say the word.

Regarding Johnson's Criteria, I believe that these should only be used in conjunction with night vision (thermal) cameras and they describe the ability to, for example, tell the difference between enemy or friend - thus, these criteria creditor only for military use!

Jan, here is the FBI's recommendations. The one I know of is from 2004/2005 and is woefully out of date with current technology.

Ok John, thanks, I saw that it was now as web text, but I had the "original" as a pdf. Looks, however, that the content is the same.
I agree - these recommendations have significant shortcomings today. But as I mentioned, it's probably the case that many of the "new" recommendations today are sometimes a mix of these recommendations and other (maby European) Standards.

Hey guys, would anyone like to offer feedback on this application note that we will publish soon?

ftp://ftp.vumii.com/Documentation/Vumii_Camera_Performance_Criteria.pdf

Roy, as your note points out, the Johnson criteria has been around for a long time.

The main challenge is that it seems military use / thermal manufacturers define 'detection, recognition and identification' differently than the commercial surveillance world.

For instance, I think most people in video surveillance associate 'identification' with being able to tell whom a specific person is ("Oh that's John, not Roy") whereas the Johnson criteria associates 'identification' with seeing that a subject is there, but not whom the person specifically is.

I don't think either is right or wrong but since the same words are being used to mean different things, there is risk for confusion.

What is the definition of "Classification"? The same as "Recognition"?

Never mind, I think I find one in http://resource.boschsecurity.com/documents/WhitePaper_enUS_2233028875.pdf .

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

Axis 4K vs. 5MP Camera Shootout (Q3518-LVE + Q3517-LVE) on Oct 04, 2018
Axis has released the 4K counterpart to their 5MP Q3517-LVE (see our test). Axis claims of both these cameras: Simply put, AXIS Q3518(17)-LVE...
AHJ / Authority Having Jurisdiction Tutorial on Sep 27, 2018
One of the most powerful yet often underappreciated characters in all of physical security is the Authority Having Jurisdiction (AHJ). Often,...
4MP Camera Shootout - Axis, Dahua, DW, Hanwha, Hikvision, Uniview, Vivotek on Sep 24, 2018
4MP usage continues to climb, especially for low cost fixed lens models. To see who was best, we bought and tested seven 4MP models from Axis,...
Amazon Ring Spotlight Cam Tested on Sep 17, 2018
Amazon's Ring has released their latest camera entry, the Spotlight Cam, which we bought and tested in our Consumer IP Camera Analytics...
IP Camera Cable Labeling Guide on Sep 14, 2018
Labeling cables can save a lot of money and headaches. While it is easy to overlook, taking time to label runs during installation significantly...
Vivotek 4MP Camera Tested (FD8379-HV) on Sep 13, 2018
Next in our series of updated 4MP testing, we bought and tested Vivotek's FD8379-HV, and entry level 4MP model claiming "top-notch quality video in...
VMS Export Shootout - Avigilon, Dahua, Exacq, Genetec, Hikvision, Milestone on Sep 13, 2018
When crimes, accidents or problems occur, exporting video from one's video surveillance system is critical to proving incidents. But who does it...
Dahua Low-Cost 4MP Camera Tested (N44CL52) on Sep 10, 2018
4MP use continues to increase, especially in low-cost models, according to integrators in our 2018 Resolution Usage Statistics. We bought Dahua's...
Drain Wire For Access Control Reader Tutorial on Sep 04, 2018
An easy-to-miss cabling specification plays a key role in access control, yet it is commonly ignored. The drain wire offers protection for readers...
Exit Devices For Access Control Tutorial on Aug 28, 2018
Exit Devices, also called 'Panic Bars' or 'Crash Bars' are required by safety codes the world over, and become integral parts of electronic access...

Most Recent Industry Reports

Hanwha Dual Imager Dome Camera Tested (PNM-7000VD) on Oct 18, 2018
Hanwha has introduced their first dual-imager model, the PNM-7000VD, a twin 1080p model featuring independently positionable sensors and a snap-in...
Camera Height / Blind Spot Added to IPVM Camera Calculator on Oct 18, 2018
IPVM has added camera height and blind spot estimation to the Camera Calculator. This is especially helpful for those who need to mount cameras up...
Axis Strong US Growth, Flat EMEA - Q3 2018 Financials on Oct 18, 2018
This spring, Axis had its best financials in many years (see Axis Strong Q2 2018 Results). However, over the summer, Axis had many products sold...
Best Alternatives to Banned Dahua and Hikvision on Oct 17, 2018
With the US government ban and a growing number of users banning Dahua and Hikvision, one key question is what to use for low cost? While Dahua and...
Video Quality / Compression Tutorial on Oct 17, 2018
While CODECs, like H.264, H.265, and MJPEG, get a lot of attention, a camera's 'quality' or compression setting has a big impact on overall...
Knightscope Winning Investors, Struggling With Growth on Oct 16, 2018
While Knightscope's new financials show the company only winning 11 new customers in the past 12 months, the company continues to win new...
Integrator Laptop Guide on Oct 16, 2018
This 18-page guide provides guidance and statistics about integrator laptop use. 150 integrators explained to IPVM in detail about their laptops,...
Huawei Admits AI "Bubble" on Oct 16, 2018
A fascinating article from the Chinese government's Global Times: Huawei’s AI ambition to reshape industries. While the Global Times talks about...
ADI's Financials Revealed + W-Box Growth Priority on Oct 15, 2018
  ADI is one of the most powerful distributors in the security industry but how big are they? How much profit do they make? How much do they sell...
Dahua Face Recognition Camera Tested on Oct 15, 2018
Dahua has been one of the industry's most vocal proponents of the value that AI creates: As part of this, Dahua has released a facial...

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