Operating a device without touching it. That's nothing new, it typically involves expensive technology and high energy consumption. At European Microwave Week 2017, an event running from 8 to 13 October in the German city of Nuremberg, TNO presents a simpler approach. Five facts about movement recognition using deep learning.
Did you know that radar can sometimes be better than a camera?
Manufacturers of consumer products are making it ever easier to operate their devices. A gaming console uses a camera to track movements and a handsfree speaker in the lounge listens to commands. While this all works well, the consumer may feel it constitutes a breach of privacy. Moreover, the camera or sensor must 'watch you' through an aperture in the device. With radar, there's no need for that: like the antenna in a mobile telephone, it is tucked away inside the device, out of sight. And the fewer buttons in the design, the better. That's the current trend.
“The deep learning algorithm has been inspired by the human brain: the more you train it, the better it performs”
Did you know that deep learning is already widely used?
TNO is investigating how far deep learning can be applied to movement recognition using radar, at home and at work. The deep learning algorithm has been inspired by the human brain: the more you train it, the better it performs. The technique is already being widely used, for example for facial recognition and to personalise social media. Its applications also include defence purposes, such as the identification of a hostile target.
Did you know that the radar already recognises four commands?
Movement recognition using radar will soon enable the consumer to operate a device without having to touch it, and the functionality will be both cheap and energy-efficient. Before the radar can recognise such movements, it must be fed with data. Researcher and innovator Bastiaan Dekker of TNO duly spent days with the radar recording gestures. The radar now recognises four commands. Beckoning is the right gesture to switch on a device. Wiping or swiping is used to zap - when watching television. Waving can be used to switch off a device. The fourth command that Bastiaan has so far taught the radar is the most complicated: do nothing. After all, if the user walks past or reaches for a cup of coffee, the device must understand that no action is expected.
“Its applications also include defence purposes, such as the identification of a hostile target”
Did you know that TNO has everything in-house needed to improve the technique?
The radar is now very good at recognising Bastiaan's gestures. When it comes to other gestures, ones it has not been trained for, the radar is less good. Broadening its range is an important activity that we will be focusing on in the coming period. In addition, we are working on the chip itself. To show that the idea was practicable, we used an existing chip previously designed at TNO. It offers the big advantage of using little power. In a next step we will research how the required performance can be delivered at the lowest possible cost. We will combine the areas of expertise this requires: we have decades of experience in building the entire radar chain, consisting of the antenna, all the hardware and the software algorithms that extract the right information from the signal.
Did you know that in the future movement recognition will have its own niche?
In the future movement recognition will occupy its own niche among other sensors. Just think of energy-efficient applications in the home, such as reviving a device from standby mode. Or the safe operation of a car radio. Combinations of sensors are also expected to increase. The kind of information radar provides is different from a camera: a camera shows, for example, the angle from which another road user approaches, while radar shows the distance to that vehicle and its speed. The challenge that TNO is now taking up is to choose the best topology for a neural network and always come up with an integrated system that works.
“Movement recognition using radar will soon enable the consumer to operate a device without having to touch it, and the functionality will be both cheap and energy-efficient”