A team from Stanford University has developed a new AI camera it claims could revolutionise the fledgling autonomous car industry.
Trying to replicate the driving abilities of a human being within the limited space of a four-wheeled motor vehicle is an almighty challenge.
Due to the sheer computing power involved and the myriad of sensors needed, developers have been left needing to connect bulky computer systems to one another making it less than ideal.
This is largely down to the fact that the computing power needed to run artificially intelligent (AI) algorithms make devices too large.
Now however, a paper published to Scientific Reports by a Stanford University research team claims it has developed something which could shrink the technology down, while being much more powerful than existing systems.
A hybrid system
This new AI camera system, the team said, can classify images faster and more energy efficiently than anything today, and could soon be small enough to embed into devices themselves.
To do this, the researchers combined two types of computer into one, creating a hybrid optical-electrical computer specifically designed for image analysis.
The first of its layers physically pre-processes image data by naturally filtering light as it passes through custom optics, entirely without a power source.
This saves the hybrid system a lot of time and energy as otherwise an electronic computer would need to crunch the numbers, requiring a lot of energy.
Essentially, the maths of AI has been outsourced to the optics of the camera resulting in substantially fewer calculations and a faster system.
Needs to shrink
In terms of speed and accuracy, the prototype rivals existing electronic-only computing processors that are programmed to perform the same calculations, but with substantial computational cost savings.
During testing, its camera was shown to successfully identify a number of different objects, including aircraft, cars and animals using natural image settings.
While the current prototype is the size of a table, the research expect that it could one day be fit into a device as small handheld video camera or an aerial drone.
“Some future version of our system would be especially useful in rapid decision-making applications, like autonomous vehicles,” said Gordon Wetzstein who led the research.
The next step in the research is to see how the camera’s optical component can be made to do even more of the AI’s pre-processing.