https://finance.yahoo.com/m/b64b239f-2cdd-3df3-ae90-aaff66a7…
Nvidia shares will rise as the chipmaker benefits from governments using artificial intelligence technology to monitor its citizens, according to a Wall Street firm.
UBS raised its price target to $285 from $266 for Nvidia shares, citing the company’s leadership in the machine learning chip and software markets.
Nvidia offers its “Metropolis” platform, which enables governments to use its A.I. technology to analyze video feeds. The offering helps cities provide traffic monitoring, law enforcement and public safety services.
Now if the cameras actually have a “brain” built in that can recognize behaviors preceding a crime for example then the motors that controls where the cameras point, focus, and track can be done in real-time which leads to more effective surveillance. The cameras brain can also then make decisions to communicate to other cameras, alert authorities, etc. This would be one use case example of how NVDA’s brains will be deployed at the edge which I believe will ultimately be much bigger than NVDA’s datacenter opportunity.
Chris
7 Likes
Chris,
Do you have any idea of what an ominous picture you paint? How do you define “behaviors preceding a crime”? How do you insure it’s only criminal behaviour being surveilled?
Yeah, I’m long NVDA as well, but this is scary stuff. I read somewhere that the Japanese police had apprehended a person wanted for some crime by picking his face out of crowd via facial recognition s/w. Sent a little shiver down my back.
6 Likes
How do you define “behaviors preceding a crime”?
You get 3 Precogs and put them in a pool (http://minorityreport.wikia.com/wiki/Precogs ), of course. ;^)
I read somewhere that the Japanese police had apprehended a person wanted for some crime by picking his face out of crowd via facial recognition s/w.
This has happened more than once in China. Heck, in Shenzhen they publicly scold you on big outside TVs if you jaywalk, and will eventually ticket you (https://www.independent.co.uk/news/world/asia/china-police-f… ).
Back to the original request to have this be edge computing - it seems far more likely that facial and intent recognition would be done in the data center on images sent from dumb cameras.
2 Likes
Somehow posting an example from Shenzhen does not boost my comfort levels about potential abuse of a system of this sort . . . And I spend a lot of time in China as my is Chinese.
Smorgasbord1,
“it seems far more likely that facial and intent recognition would be done in the data center on images sent from dumb cameras”
Technology seems to oscillate between central processing and distributed processing. Today’s surveillance networks can consist of millions of cameras, making centralized real-time analytics across the network unrealistic. Processors (hopefully NVDA
have now become powerful enough to remotely process the primary analytics at each camera and, if a suspicious event takes place, forward the information immediately to a screen at the command center.
For those who like to read:
https://www.theverge.com/2018/1/23/16907238/artificial-intel…
https://www.prnewswire.com/news-releases/global-network-came…
https://business.panasonic.com.au/security-solutions/sites/d…
Jeff
2 Likes
“it seems far more likely that facial and intent recognition would be done in the data center on images sent from dumb cameras”
Technology seems to oscillate between central processing and distributed processing. Today’s surveillance networks can consist of millions of cameras, making centralized real-time analytics across the network unrealistic.
There are many products like this (just one example):
https://www.ambarella.com/products/security-ip-cameras
Ambarella (became a significant camera IP company by providing tech to GoPro) makes camera sensors with the ability to run deep neural networks right on the camera chip, minimizing bandwidth within the camera as well as from the camera to the cloud.
From a privacy point of view this kind of architecture may become required…no images are stored or transmitted, just hits on what it was trained for, such as certain behaviors.
Mike
Mike,
As part of my speckled past, one of the firms I owned was involved in mid-sized (the largest was about 50,000 cameras) surveillance networks. That was seven years ago and technology has moved on - but privacy concerns haven’t. While there are, indeed, places where privacy has to be involved (either legally or due to trade union pressure), such as classrooms, locker rooms and toilets, one of the functions of these networks is to record data for retroactive forensics.
The problem is that a small handful of mere mortals can’t watch even 100 screens simultaneously, let alone numbers many orders of magnitude greater. Additionally running real-time analytics on all these video feeds would require head-end hardware costing beyond the budget of most entities. While all the data could still be transmitted to the storage units on a real-time basis, today it’s found more cost effective to do the initial processing for alarm consideration at the camera itself and send that alert as an overlay with the video feed, thereby prioritizing what the human security personnel see.
Most of the camera companies are Asian (with Japan associated with the higher end ones), though Phillip and Bosch are European. How much of the mind-share of the camera designers Nvidia can grab - I just don’t know. The camera designs are retained way past the date when others would update technology as this market is not pushing for bleeding edge technology, but rather for records of long-term reliability of a model in the field. That said, believe it or not, it’s neophyte politicians who push for change as they don’t like to be associated with “old technology” if the movies they watch show what “can be done”.
Jeff
2 Likes