In my past life with Intersil, I had occasion to show customers an eval board that was set up to do gesture recognition. It used some of Intersil proximity sensors. It consisted of an array of sensors arranged to detect whether a hand was waved left-to-right or up-and-down. With just that two-axes information, you could arrange to control a variety of devices. Quite useful in automotive applications, though I'm not sure it was ever pursued beyond the eval board stage.
Recently, I saw a video from the UC Berkeley school of engineering that showed the work the folks there were doing on gesture recognition using ultrasonics. They claim that their technology is far lower power than the typical optical measuring methods. They compared their version to a camera and claimed 400μW vs. 1W power draw. The prox-sensor method that I cited above doesn't need 1W, but it's still rather power hungry, so this ultrasonic version is of interest for portable, battery-operated devices.
From promotional material from the UC Berkeley, there is this background info:
Researchers at the University of California, Berkeley have developed an ultrasonic graphic recognition system that can track a user's movements and translate them into inputs to an electronic device. Their ultrasonic 3D range sensor system uses batch-fabricated micromachined aluminum nitride ultrasonic transducer arrays and custom CMOS electronics. This technology may be useful in the development of practical gesture-controlled computer interfaces. Optical 3D imagers for gesture recognition, such as Microsoft Kinect, suffer from large size and high power consumption. Their performance depends on ambient illumination and they generally cannot operate in sunlight. Ultrasonic gesture recognition systems measure sound waves to turn gestures into inputs, allowing for comparable performance to optical systems but with less power consumption and environmental restrictions.
This method of sensing should lend itself to large-scale integration. If we could combine the ultrasonic sensor array, the transmitter driver, the receiver pre-amps, and the processing circuitry onto a chip, we'd have a very useful device. Conceivably, it may not be practical to put the entire ultrasonic sensor array on an IC. They may require too much area and may need to be spaced too far apart. The alternative is to: a) keep the transducers completely separate from the circuitry described above, but with all the circuitry still integrated; or b) put just the transmit driver, the receiver preamp, and one ultrasonic transducer on an IC. This is still a useful amount of integration to squeeze on one piece of silicon.
Have you worked on any similar gesture recognition devices? What problems did you have to deal with?