Foolish Strategy: OEMing Facial Recognition

By John Honovich, Published Dec 13, 2018, 12:44pm EST

Almost as 'hot' as face recognition marketing right now is OEMing facial recognition.

Last year, they were a who's who of company's with backdoors, now they are world leading computer vision experts.

The strategy now though is OEM face recognition, sell appliances through distribution for hundreds of dollars, profit fail.

Here is an example from one enterprising American entrepreneur's facial recognition system:

Actually, it is not. Not only are the OEMing the product, they lazily blurred out the name of their Chinese OEM in their supposedly own marketing, as shown in the original image below:

That company is not alone. Every week we see more and more companies who cannot spell CV release facial recognition systems.

Why ** ****

***'* ***** **** *** bull **** ** *** a ******* ****** *** facial ***********:

  • **'* * *** *****. It's **** ** **** into *** ******.
  • ***** *********** *** ***** it. **** ** *** want ** ** **** behind.
  • ****** *********** *** ******* criminals, ******* **** *** more (*************)
  • **** ******** ****** **** cameras *** ********* (*******, sell ** ** ******, profit) ** ** ***** work *** ****** ***********.

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

*** **** **** *** tech ***** ******* ************ works ****** ** **** of *** ******* *** not *** **** ***. Dahua *** ******** ** thousands ** '*********' ******* on **** ******** **** *********** ****** works ****** (******) ***** ****** *********.

*** **** ** ** worked ******, *** ******** is ********** ** ******** problems:

  • ******* ********: *** ***********, low-cost ************ ** ***** on *** ******* ** low *******. **** *** box, **** **** *****, move **. ****** *********** does *** **** **** way. **** **** **** scenario, ****** *********** *********** is *** **** ********* to ****** *********** *** lighting, ***** ******** ****, knowledge *** ***** ****** cannot ** **** ******* significant ******* ** ****. Support ***** *** ***** to ********* ******** ***** customer satisfaction **** *******.
  • ******** ********: **** ** cameras, ************** ****** *** happy ****** **** *** images **** ***. **** if **** *** *** perfect, **** *** ********* more **** **** ****** with *** ******* '*****' outcomes. *** **** ****** recognition, **** *** ****** is *****, ** ** clearly *****. *** **** matched ** *******, *** gets ******* ** ***, etc. ***** ****** *** that *** ****** ***, at ****, ** **** confidence, ** *****. **** results ** **** ******* problems, ** *** ******** abandonment ** *** *************.
  • *********** ********: **** ** these **** / ************ ** when ***** **** ***** problems? **** *** **** do? **** ** *** know ******** ***** ****** recognition *** ** **** have ****** ** *** source **** ** ************ to *******. *** *** companies '**********' **** *** likely ** * ******* position, ********** **** ****** or ****** *********.

****** *********** *** ******, done ****, **** ****** support, *** *** ******** of ******* ****** *********** into *** **** *********, box-moving, ****** ** * recipe *** *******.

Comments (22)

 Do you have a rough list of companies doing this, and/or which products are being OEMed most often? Trying to gauge the "hot" reference.

We don't have a directory of facial recognition providers but we will put that in queue. 

As for 'hot', I've seen this at numerous shows this year, online marketing, etc., and it's the most visible video surveillance shift in marketing in the past year.

It's the only new thing our boring industry has seen in years and years and years... Sure, 3MP went to 4MP and is going to 5MP now... number +1 isn't innovation though.

Seems to me that many of your complaints are actually the fault of the "Professional Installers" whom don't know very much about the physics of light and how to create a "good" image... You can buy a $3000 Nikon DSLR and if you can't take a good picture you aren't going blame Nikon...

The problem is how it is marketed and perceived, again the #1 problem in our industry is educating the end-user and/or installer. If used properly I don't see why there's such a big problem with facial recognition/detection. It could even be useful if one day it can tie into the "Most Wanted" criminal databases.

These companies are releasing products that the factories/manufacturers are making... Should they say "Wait, no... we don't want that feature in our NVR! It's a hot-topic bandwagon marketing ploy! It only works when you set it up properly! TAKE IT OUT!" ?! (We took out P2P for security reasons in one of our lines much to our detriment...)

Our industry is a commodity... everything in this industry has been commoditized by the Chinese-American distributor's Race to the Bottom- this is also the reason InvidTech blurred TVT's name out... to protect their market-share from basement-bargain-budget "distributors"- and you just destroyed what little margins they could achieve...

They'll do it with FR as well, because this industry is controlled by one, and only one, chipset maker (Chinese owned too!). Hik gets FR the same time Dahua gets FR the same time TVT gets FR... and so on. We're all handcuffed and crippled in this industry. If only we could all remember : You're not in Shenzhen anymore!!!

#1 problem in our industry is educating the end-user and/or installer. 

Rule of thumb: 'Education' problems are product problems that companies do not want to address.

This excuse has been going on for years, see this 2010 IPVM post Why Education is Not Enough for Video Analytics. Nearly a decade later, it's still the same false excuse.

If used properly I don't see why there's such a big problem with facial recognition/detection.

What is 'used properly'? Confining it to perfect camera angles and lighting and low user counts? Users want it to work in their conditions, not to have to 'educate' themselves how they need to warp their usage and environment.

They'll do it with FR as well, because this industry is controlled by one, and only one, chipset maker (Chinese owned too!)...

Hik gets FR the same time Dahua gets FR the same time TVT gets FR... and so on. We're all handcuffed and crippled in this industry.

This is simply wrong. Huawei, among many are trying to offer this but facial recognition has various ways to be implemented (think NVIDIA, Intel, Ambarella, others) and it is not controlled by Huawei.

What is 'used properly'? Confining it to perfect camera angles and lighting and low user counts? Users want it to work in their conditions, not to have to 'educate' themselves how they need to warp their usage and environment.

It would be the "proper" scenario for the ever-cheaper IP Cameras to work with FR. Something like access control, you'd want facial recognition set up at entry-ways at proper angles with proper lighting. I want a flying car... but I have to use the technology I have. Users need to be educated and adaptive until the technology catches up with their "wants".

This is simply wrong. Huawei, among many are trying to offer this but facial recognition has various ways to be implemented (think NVIDIA, Intel, Ambarella, others) and it is not controlled by Huawei.

It's actually not wrong. At all. Half of your article today is complaining about this "Strategy" and Huawei = HiSilicon and every Chinese MFG uses HiSilicon so whatever features HS adds to their chipsets and SDK will trickle down to the manufacturers. Make a timeline of when each company offered FR, and look at the chipset in each of these units. Look at the price of the units with this technology.

If you're reaching outside of our industry, sure, better- more elevated companies will do better things with "Facial Recognition" and maybe it won't be commoditized in THOSE industries. In terms of CCTV? I'm 100% right.

If you're reaching outside of our industry, sure, better- more elevated companies will do better things with "Facial Recognition" 

Ambarella is outside of 'our industry'? Your part of the industry may be different than the mid to high end but Ambarella and now Openvino, which various major manufacturers are developing on, is certainly part of the industry.

In terms of CCTV? I'm 100% right.

We'll see. My bet is that these ~$500 face recognition DVRs are going to bomb. And the fact that you are betting on 'educating' the user on 'proper' usage means the odds are heavily in my favor.

I wasn't talking about Ambarella... I was talking about HiSilicon and I'm not betting a dime on educating the user. I know that will fail. I think  we're trying to say the same thing and getting confused somewhere along the way. My whole point is China is commoditizing what could be a cool technology. It's dead on arrival, there is no strategy.

My whole point is China is commoditizing what could be a cool technology.

It's only a 'commodity' if it 'works.' Commodity HD 'works'. Commodity facial recognition does not work, ergo my position.

Again, it does work. Use it how it is meant to be used... Do you expect a bio-metric fingerprint reader to work if you're sliding your finger around a few CM above the reading mechanism?

Again, it does work. Use it how it is meant to be used... 

Users want to use it in the way they want, not how it is 'meant' to be used. 

What do you want to do? Tell people - you need to make sure people look at the camera, you need the camera to be mounted 5 feet high, you cannot put it near sunlight, people cannot wear hats, etc. Most people are not going to substantially change their practice to conform to the limitations of the product.

Well... Like I said, we'd install it in the way that it works best. Entrance points, non-extreme angles, well lit rooms... Same way I'd install a fingerprint reader to read fingerprints in the environment in which it can do so successfully. We're in the infancy of the technology in our lagging industry. We're not living in "Minority Report" just yet.

I feel like you're making a mountain out of a mole-hill with the "it doesn't work!" claims. It does work. Most of your claims I haven't even encountered as making it "not work". What happens if you plug a non-IR camera into a dark room? You won't get an image... "But I WANT an image!" sounds so childish.

We'll just have to agree to disagree I guess, because FR is really such a non-issue in our industry. Give it 5 more years to catch up with modern technology. Our industry is always behind the real world.

Lol... what if we applied your logic of end users to cars? But I WANT to drive on water! I don't CARE that it's meant to be on asphalt! Can't you see how ridiculous that is?

Facial recognition is definitely one of the most challenging services in video. You can't just slap up a camera and expect it to work; IPVM tests prove that. It is an engineered system that does take support to tune, and it doesn't work everywhere.

Unique use of FR: https://www.theguardian.com/music/2018/dec/13/taylor-swift-facial-recognition-stalkers-rose-bowl-concert

The reality is while nothing is plug and play face recognition is the complete opposite end of the spectrum. 

This was a similar fate with video analytics where the calibration is a major undertaking. There are innumerable factors in the setup for an analytic camera and without the experience in knowing what to do it becomes useless. 

Face recognition is not a biometric in the sense of a fingerprint and a guarantee of accuracy. If you look at most of the face recognition solutions they return responses in % of probabilities. 

When you use your iPhone and unlock from 12” - 18” away is very different than a camera on a wall at a distance and angle where it’s not subject compliant. 

Fine tuning an analytic or face recognition cost more in technician hours then the cost of these entry level machines. This especially when calibration may require several visits to address varying environmental conditions. 

Regardless, no question there are differences in the quality of the algorithms as can be seen from NIST testing

This was a similar fate with video analytics where the calibration is a major undertaking.

The analogy of face rec to video analytics is more appropriate than face rec to resolution. Lowering cost for higher resolution works since image quality 'just works' in that most people will accept it as is, but just like low-cost vendors have marketed video analytics for years with poor adoption, same will happen with facial recognition. The major difference is that the push for facial recognition is so prominent where video analytics have generally been a minor selling point for low cost systems.

[IPVM Note: Poster is from TVT.]

Every new technology would bring a lot disputation, there are people appreciate it and get excited, some only see the concerns, but the disputation is the one really push the technology being improved and moving forward.

Look back to 2012 when the IP technology started getting into the market, lots people concerned about the hassle, the education, the techsupport, look what happen today, more than 70% are IP in US market.

The time HIKVISION started flying in the US market is 2013, full line of IP products is the key reason of their fly. (possibly HIKVISION is not a so positive example today in the US market, but what i am trying to say is that the new technology bring new opportunity, the one who are more open to it and embrace it would have more chance to benefit from it)

There are different levels of FR products, for the applications like airport, Metro Station, Stadium etc which have huge people traffic, the FR products need to have the ability to detect and recognize hundreds faces at the same time, NVIDIA based FR products are the idea option. But for the applications like Shopping Store, Office, Gas Station, Hotel, Resturant, Hospital etc which do not have that much people traffic, the NVIDIA based FR products would be too good and would not be affordable.

Many comapny are marketing the FR products today, not everyone could deliver a good one which could frustrate the market a bit, but there are people who have been dedicating into FR and be able to deliver a decent one, which could definitely bring a lot value to the industry.

IPVM may want to keep updating and test more FR products from different companies.  

"what i am trying to say is that the new technology bring new opportunity, the one who are more open to it and embrace it would have more chance to benefit from it"

This is false logic.

You can willfully ignore the value of skepticism in product development, but that does not make your position valid.

'more open to it and embrace it' implies that those who criticize the current inadequacies of a technology play no role in making the product more beneficial - and only 'believers' move this forward.

I think exactly the opposite is true - it is the skeptics that hold the 'believers' feet to the fire.... requiring the 'believers' to deliver products that provide actual value to customers who buy the product - instead of products that (in their current iterations) under-perform in real world scenarios.

Believers and skeptics should be on the same team, imo.

 

#4, thanks for your feedback. I appreciate it.

Look back to 2012 when the IP technology started getting into the market, lots people concerned about the hassle, the education, the techsupport, look what happen today, more than 70% are IP in US market.

That's factually wrong and importantly so.

It was clear even back in 2008, that IP / MP was going mainstream. Here is one of IPVM's first posts back in 2008 that accurately captured the shift: H.264 Makes Megapixel Go Mainstream (Arecont Vision). And back in 2011, we again accurately reported Why The Market Has Already Tipped to IP.

This is important because IP never faced the types of problems that analytics and facial recognition have (and continue to have). Even for people who did not know IP networks, once it was set up, it 'worked' and for the most important criteria for users (image quality), it worked better.

This is not the case with facial recognition. It still makes frequent mistakes that most users can neither understand nor tolerate, especially at the level / approach manufacturers like yourselves are using (OEMing to small relabellers who sell cheaply through distribution).

new technology bring new opportunity, the one who are more open to it and embrace it would have more chance to benefit from it

I actually agree with you here. The risk though is that the new technology is not ready / mature. Companies like Object Video and 3VR (who combined took in over $100 million in VC funding) embraced new technology and new opportunity but they did it before the technology was good enough for customers to embrace and they suffered the consequences for it.

The technology overall is better today but the technology is almost certainly not good enough especially the way you are selling it.

We will certainly continue to test facial recognition, analytics, and other emerging technology. Thanks again.

Looks legit... =p

John - There are a number of great points in this article but also a bunch of things that are not quite clear or factually correct. Let's start with your premise that OEM relationships between hardware/software manufactures and face recognition (FR) companies are a poor strategy. I think you should be careful to define which purveyors of FR you are calling out as "who's who of company's with backdoors, now they are world leading computer vision experts." I believe you are referring to the likes Hikvision, Dahua, Huawei and other Chinese manufacturers but you paint our entire industry with a broad brush.

Our company, CyberExtruder, is one of the most senior providers of FR software in the industry and has a go to market strategy that includes OEMing our software to partners wanting to enable their products with state-of-the-art face matching capabilities. There is a world of difference between the products we offer and the crap that is bundled into these low end NVRs. For anyone not familiar, real FR - that is to say a face matcher that actually works in real world deployments, doesn't run on these low power imitation computers. Real FR requires an actual (dedicated) computer to perform it's job.

We wholeheartedly agree with your statement "The fact that the tech being offered commercially works poorly is part of the problem but not the only one. Dahua has hundreds or thousands of 'engineers' working on this and their face recognition camera works poorly (tested) under modest standards." The greater truth is that companies like Dahua and Hikvision are not really interested in developing a commercial product that works well - they're simply interested in offering something (won't even call it a product) that they can label as 'face recognition' for the sake of ticking the box on the hot trend for the sake of increasing top line sales. 

You are generally correct when you say "Even best case scenario, facial recognition performance is far more sensitive to camera positioning and lighting, which requires time, knowledge and often simply cannot be done without significant changes or cost."  However, there are exceptions (insert shameless plug here), and face matching that is fast, accurate and scalable really does exist in an affordable package suitable for a variety of use cases.

I encourage you and IPVM to look at FR offerings in the same impartial and even-handed manner in which you evaluate other products. You are shining an bright light on an important topic - one in which the OEMs and resellers need clear  guidance so that they don't waste their time and customer's money while inadvertently burning their reputations. 

Not everyone who provides FR is "leveraging open source or shared libraries." Some of us have been in the business for a long time and have significant bona fides to support our claims and really look forward to IPVM style shootouts.

Jack Ives, Chief Operating Officer
CyberExtruder

That oem company you show is actually owned by a Taiwanese owner buying from Chinese manufacturers and stating the product is exclusively theirs. Facial cameras are not very good and never will be as they are never accurate

Why does hik, dahua, tvt and uniview want to compete with themselves?  All manufacturers should stop doing oem and stop driving prices in the cellar. 

Read this IPVM report for free.

This article is part of IPVM's 6,738 reports, 909 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