Stay ahead of the pack
with the latest independent smart building research and thought leadership.
Have an account? Login
Subscribe Now for just $200 per year per user (just $17 USD per month) for Access to Quality Independent Smart Building Research & Analysis!
What Exactly Do you Get?
- Access to Website Articles and Notes. Unlimited Access to the Library of over 1,700 Articles Spanning 10 Years.
- 10% discount on ALL Memoori Research reports for Subscribers! So if you only buy ONE report you will get your subscription fee back!
- Industry-leading Analysis Every Week, Direct to your Inbox.
- AND Cancel at any time
The simple truth is that intelligent computer vision is an immensely difficult challenge to take on. But with video cameras now delivering better image quality, enhanced processing power and improved software algorithms, can we expect cameras to start delivering "real" intelligence?
It was in 2011, when the IBM Watson computer won the US game show Jeopardy, beating 2 human champions.
For that, IBM developed DeepQA, a massively parallel software architecture that examined natural language content in both the clues set by Jeopardy and in Watson's own stored data, along with looking into the structured information it holds. It took about 20 researchers 3 years to reach a level where it could participate in a quiz show performance and beat human opponents.
Contrast this with Google's image recognition efforts in 2012, when they connected 16,000 computer processors in a neural network that taught itself to recognise cats from thumbnail images extracted from millions of Youtube videos. This was in itself a significant breakthrough because the algorithm was given no help in identifying features. Currently most commercial machine vision technology is performed by having humans “supervise” the learning process by "labelling" specific features.
While computers can now out perform humans in tasks like natural language processing, they are still a long long way from matching our brains at image processing.Next Tuesday, 15th September, we have holding a FREE Webinar with Carter Maslan, Founder & CEO of Camio and previously Director of Product Management at Google. Camio is a startup providing home video monitoring, and they are currently expanding their use of artificial neural networks — a machine learning technique that draws on the way networks of neurons in the brain adapt to new information. Currently they determine that something being recorded by a user’s camera is interesting by detecting a significant amount of motion in a scene. Maslan says the company uses neural networks with each video camera that concurrently vote on what they think the user would consider interesting. The technology is proved right or wrong based on videos the user eventually opens, plays, and deletes. Home Security is not the only vertical benefitting from more advanced machine learning. Analytics are proving useful in niche areas like casino surveillance, where video cameras are monitoring for customers who exhibit specific behaviour patterns, like card counters. As analytic algorithms improve, casino surveillance will be able to implement systems such as facial recognition to identify subjects and cross reference them with stored "Big Data". In the medium to short term, the future for Video Analytics in security seems to be within functions that have market specific rules. Among these will be Retail, Transport and Business Intelligence applications.
Large scale deployment of truly intelligent analytic technology across all market sectors, will require breakthroughs in image recognition algorithms; perhaps even fundamental changes in the way software is designed. This will NOT come from the security industry. But in the meantime, incremental improvements in camera image technology and processing power will move us towards better market specific intelligent video.
