Camera corner: Deep in thought
October 17, 2019 By Colin Bodbyl
By now everyone has heard the term AI.
Some believe it will drive the next technology revolution, while others fear it will bring the end of the world, but what is AI, anyway? Artificial Intelligence (AI) is a machine or computer performing a task that would normally require human intelligence to complete. AI has existed since the 1950’s when Alan Turing first contemplated whether machines could think. While artificial intelligence has existed for a long time, it’s first transformational breakthrough came in the form of machine learning.
Machine learning is a subset of AI. With machine learning, a human manually trains the computer on what criteria it needs to use in order to perform a given task like identifying an object in an image. For example, with machine learning, a human may tell the computer that anything with four legs, a tail, fur, pointed ears, and two eyes is a cat. This same computer could now search through thousands of images using machine learning to classify cats versus other animals based on the set criteria. As the computer returns false results a human could interpret what features the computer is struggling with and further refine its training in order to improve accuracy.
The challenge with machine learning is that accuracy levels often fall well below human levels. While most humans could identify the difference between a cat and a different animal with near 100 per cent accuracy, machine learning would peak at around 80 per cent. Ultimately, it was another breakthrough in AI, called deep learning, that changed all this and fueled the recent proliferation of AI.
Deep learning is a subset of machine learning. Where machine learning requires a human to pre-program criteria into the computer, deep learning takes a different approach.
Rather than telling the computer what a cat looks like and then asking it to identify cats based on those criteria, deep learning is programmed by telling the computer that you are about to show it several thousand images of cats, and then allowing the computer to analyze each image and decide for itself what qualifies each one as a cat. By allowing the computer to identify key traits across images, deep learning is not only more accurate than machine learning, but also continues to get more accurate with every new image it sees.
It is the ability to continuously self-improve without human intervention that has brought AI back into the spotlight. Outside of the security industry, AI has already outperformed humans in fields like medicine where it can more accurately identify problems on an x-ray or CT scan. In this case the AI also benefits from infinite memory and the ability to learn every possible condition that could appear on a given scan, where traditional doctors cannot retain that same knowledge.
In security, AI is mostly being used to identify specific objects in live and recorded video, but it will eventually do so much more. Understanding how AI works at its most basic level is critical to grasping its potential. In the past, we have seen innovators in our industry compete to have the highest resolution camera, or the best video compression. AI is bringing the next technology race to the security industry but this time the finish line is truly unknown.
Colin Bodbyl is the chief technology officer of Stealth Monitoring (www.stealthmonitoring.com)
This story was featured in the October 2019 edition of SP&T News magazine.
Print this page