Member Discussion

Making List Of Deep Learning Video Surveillance Providers

IPVM is making a list of security companies that are using deep learning. Deep learning goes beyond standard pattern-matching or size-filtering video analytics to incorporate better abilities to recognize and classify objects in a variety of settings, and can often adapt to environmental or scene changes without requiring re-calibration.

Companies identified so far and their use of deep learning:

We will be doing more coverage of the deep learning trend in the security industry in the future and want to start with this list of which companies should be included.

If you know of other companies in the security industry using deep learning, please comment on the company/use or email me: bkaras@ipvm.com


Herta, who were supposed to be at ISC and weren't, use the Cognitec algorithms.

So if you include Herta, then you'll have to include Cognitec & FaceVACS

Additionally NEC and their Neoface product.

FST Biometrics is another company using Cognitec algorithms

Is Cognitec deep learning or traditional analytics?

Dare I say Giant Grey?

Did you mean OMNI AI?

Given their history, I was leaving them out until we see some indications of stability and continued product offering.

Say want you want about BRS Labs/Giant Gray/Omni AI but they pioneered the use of "deep learning" marketing terms, years before anyone even knew what they were talking about.

Not sure I agree.

VideoIQ was using the "self-learning" term back in 2008, it was even mentioned in IPVM's 2009 test of VideoIQ.

In addition, we were shipping product that was using deep learning/neural networks to far more customers than BRS ever did.

Fair enough.  

And I'm sure about the number of sales being far greater with VideoIQ.

But BRS Labs did get a significant early share of press regarding its "approach".  From a fairly in-depth article in March 2009, which names many Analytics mfrs, BRS was mentioned thusly:

Another new technology is adaptive learning surveillance software. BRS Labs developed what it calls a cognitive video analytics system that examines video for behavior patterns. 

Or winning an award at ISC in April 2009

In contrast to traditional video analytics software, such as ObjectVideo , ioimage , Lenel , BRS does not need users to draw tripwire or region of interest (ROI) and define the rules. It uses a machine learning based approach to adaptively learn what behaviors or activities are "normal" in the scene, and issues alarms when finding some anomalies.

 

Say want you want about BRS Labs/Giant Gray/Omni AI but they pioneered the use of "deep learning" marketing terms

Criticizing or praising them at this point is neither here nor there. If they demonstrate ability to ship product commercially, which there are doubts given the lawsuit and 3 company names in the past year, we are happy to include them.

I believe Intelli-Vision should be consider although their product is software that can be embedded on cameras or within a VMS or NVR .

Dave -

I don't see anything on their page that looks like they are specifically using deep learning. Do you have additional information on their technology?

 

Brian:

I did a search on their site and this is what came back:

https://www.intelli-vision.com/?s=deep+learning

 

Thanks. They should work the deep learning buzzwords into their primary product descriptions.

 

IntelliVision's algorithms are based on deep learning. You can evaluate their LPR engine integrated to AxxonNext. With low resolution images it can predict number even when human can't. Very impressive!

You could add Ipsotek to the above list.

A tighter definition / qualification for 'Deep Learning' would be useful - the majority of the above companies have been doing the same thing for 5+ years and previously used other buzzwords such as cognitive analytics. 

Not sure how ALPR fits in to this. Even the more advanced platforms with features such as convoy analysis are not really learning anything autonomously.

I left the "deep learning" qualification somewhat open, as implementations and use cases vary.  For any companies on the final list, we will do more investigation to determine how/where they are using deep learning in their products.

I disagree that the 'majority' of the companies listed have been doing the same thing for 5+ years. There are several startups (AvidBeam, Camio, Deep Science, IronYun, OpenALPR, Optigo) that did not even exist 5 years ago, and others like AgentVi that have been around, but only more recently began using deep learning in one or more products.

ALPR fits in via the companies listed using deep learning as a mechanism to train the algorithms both to recognize where license plates are in an image, and to recognize how to read characters on plates quickly/accurately, among other things.

Hi Brian,

Entropix has has been 100% deep learning neural network based and built upon NVIDIA GPU technology since before IPVM first reported on us.   Would love to get it added above.  

Added, thanks!

Thank you.  But for clarification.  We don't do "enhancement".  That's polishing and spit-shining tricks that are done on single aperture image capture.  We do multi-aperture image reconstruction.  

Enhancement is bad for computer vision analytics.

Reconstruction improves/extends range of analytics.

Major differenence in terms of ability to use in conjunction with video analytics such as facial recognition etc.

Nate

For those with the note next to them, "...in development", does that mean they have not yet released an actual product yet, or at least one that can definitely be said to be out of "beta mode".?

For Dahua and Hikvision, yes, I do not believe they have products that are yet publicly available, but both are large companies that have shown products based on deep learning.

At least for Hikvision, there are definitely deep learning products being deployed in China; not clear if or when for the rest of the world.

Thanks for the clarification. I'm just very averse to anything that might actually be more vaporware than actual product. After years of empty promises by manufacturers, I think they should only be recognized for providing technology when they have an actual product available.

Qognify VA/Suspect Search

Thanks, added Qognify to the list.

Related, Hikvision issued a release that they took first place in an OCR competition. According to the release, their LPR cameras are based on the same technology used in the competition.

 

 

Impressive! Universal OCR is something really new in our market. Interesting to see how good it works with car numbers.

IQrity Inc. - DL facial recognition SDK is available now. DL video analytics is not published yet.  

Thanks, added them in.

 

I'm happy to see that AxxonSoft supports camera embedded IQrity AnalyticVideo Tracker: http://iqrity.com/cms/product/iqrity-analyticvideo-tracker/

AxxonNext has a lot of unique features based on this tracker including MomentQuest (forensic search), TimeCompressor, AutoZoom and Tag&Track.

For those who are less familiar with deep learning, and how it can be applied to things like LPR/OCR, this new Google mini-product is a good example of using deep learning to convert "squiggles" to recognizable things:

Google AutoDraw

 

What about Kiwi?

I do not know a lot about them, but do know that Genetec is pushing this new partner and has even started Integrator Certification for Kiwi integration.

https://www.kiwisecurity.com/#

One more: http://evolvtechnology.com/ (Evolv)

Facebook's newly released Caffe2 out of UC Berkeley.

https://github.com/caffe2/caffe2/wiki/Model-Zoo

Significant area of research and advancement.  Not quite video-ready plug and play, but certainly a huge area of effort.   Shows that open source efforts may lead the way in new technologies.

 

iCetana?

Calipsa, London based Deep Learning startup. Will be exhibiting at IFSEC.