Corporate Profile
September 2002

Computer or "machine" vision is one of the oldest and most widely-researched fields in the post-industrial era, with roots going back to the dawn of television. But today, as much an art as a science, practical applications have been constrained to either extremely narrowly defined problems – such as automatically focusing cameras, or the machine assembly of pre-specified components – or to government-sponsored applications, such as mapping or spy satellites, where vast sums of money can be thrown at extremely complex and subtle technology.

The key challenge for most applications is to separate out specific objects in a scene from the background, or from one another, given only the 2-dimensional or "flat" images provided by today's camera technologies.

Both digital still cameras and video cameras utilize a common image sensing methodology in which light reflecting from objects in a three dimensional scene is projected by a lens onto a tiny, flat, semiconductor chip that contains hundreds of thousands to millions of sensing pixels arranged in a rectangular array. Each of the pixels converts light of certain frequencies – or colors – to electronic charges, which are then "read" by adjacent electronic circuitry, and each of the pixels represents a specific, unique feature in the scene. But the only information that such devices provide is the brightness and color of the light illuminating any individual pixel at the time it was read, and the relative location of each pixel.

Perceiving specific objects or other identifiable features in a scene, given only a 2-dimensional array of a million or so tiny colored dots, is a profoundly difficult problem. Thousands of mathematical algorithms have been researched in the last 60 years that attempt such identification, but many depend upon either a complex physical setup, such as multiple cameras and controlled light sources, or upon a foreknowledge of all of the elements expected in a scene. And virtually all require substantial amounts of computing power.

Canesta, Inc, an early-stage company in Santa Clara, CA, has made significant breakthroughs in this regard by its invention of a low cost electronic perception technology. The technology includes new types of chip-based image sensors, similar in size, complexity and cost to commodity-priced video camera chips, that are uniquely able to resolve the three-dimensional features of a scene relevant to specific applications. That is, rather than the brightness of a specific color of light reflecting from nearby objects being captured by each pixel, the distance from the illuminated object to each pixel is determined instead. Since each pixel in the sensor is illuminated by a different feature in the scene being viewed, the result is a true, three-dimensional representation. In addition, the sensors can operate at over 50 frames per second, making possible real-time applications, such as keystroke recognition in a projection keyboard application, or lane-change alerters in automobiles.

The resulting geometric information acquired continuously from the scene is so rich that, depending upon the application, very often much of the computation can be performed in real time by comparatively lightweight processors embedded right into the sensor chip itself. This is in contrast to contemporary systems that can take from several seconds to several minutes to compute a 3-D map of a single frame. For example, in Canesta's recently-announced Integrated Canesta Keyboard, a user's finger motions typing over a "projected" keyboard are resolved into actual keystrokes, and then into serial data, for use by an OEM device such as a cell phone, smartphone, or PDA.

The availability of technology of this power in such a tiny, low-cost format means that a wide variety of machines and electronic products will be able to perceive and react to nearby objects or individuals in real time through the medium of sight. The applications are endless.

An excellent example is the first Canesta product that utilizes electronic perception technology – the Integrated Canesta Keyboard™, announced in September, 2002. A mobile or wireless OEM device such as a smartphone, cell phone, or PDA can, for the first time, allow users true data input comparable to a full-sized mechanical keyboard. Using Canesta's technology, the image of a full-sized keyboard is projected by a tiny pattern projector onto a convenient flat surface in front of the user. Electronic perception technology, integrated right into the OEM device, is then used to resolve the user's finger movements typing on the projected keyboard into serial keystroke data immediately usable by the embedded processor in the mobile or wireless device.

Canesta was founded in April, 1999, and is located in San Jose, CA. The company has filed or has been granted in excess of 30 patents. Investment to date exceeds $20 million, from Carlyle Venture Partners, Apax Partners (formerly Patricof & Co Ventures, Inc.), JP Morgan Partners (formerly Chase Capital Partners), TechFund Capital, and Thales Corporate Ventures (formerly Thomson-CSF Ventures.) Canesta has over 40 employees.

-30-