Global Sources
EE Times-Asia
Stay in touch with EE Times Asia
EE Times-Asia > Embedded

CMOS imager chips enable true machine vision

Posted: 19 Dec 2006 ?? ?Print Version ?Bookmark and Share

Keywords:CMOS? image sensor? light sensor? Canesta? sensors?

Now that automobiles are beeping to alert you of obstacles while backing up, giving you directions on where to turn, and parallel-parking "hands free," you might have thought that machine vision had already been perfected. Such is not the case.

In fact, all of the above applications are using make-shift technologies that substitute for true machine vision. New CMOS imager chips are emerging that directly sense depth3D pixel-by-pixelenabling machine vision to realize its goal of perceiving objects, and reacting appropriately.

The automotive sensor market already tops $2.5 billion, according to ABI Research. Another $750 million is divided among security, industrial automation and videogaming uses of electronic sensors, according to Frost & Sullivan, the Automated Imaging Association and Piper Jaffray, respectively.

Distance calculation
Current automotive systems calculate distance, such as those that warn drivers of obstacle behind them as they back up, by using ultrasonic sensors. More sophisticatedexpensivesensors are under development using radar and lidar, but companies worldwide are trying to find cheaper ways to use cameras to calculate distance. Cellphones have invoked economy-of-scale enabling digital cameras to rival analog sensors in cost.

"Cameras are becoming so inexpensive that everybody in the business is trying to use them instead of more expensive ultrasonic, lidar and radar sensors," said David Alexander, senior analyst at ABI Research. "The other motivation for going to cameras is that different driver assistance applications could potentially share the same camera and even the specialized computer chips and software needed to calculate distance."

Machine vision apps
Applications that could be performed by machine vision algorithms and a 3D camera include collision avoidance, lane departure warning and lane keeping (that steers you back into your lane), reverse obstacle warning, pedestrian detection, headway monitoring (to maintain proper distance from the cars in front of you), night vision, adaptive headlight control, traffic/speed sign recognition, blind spot detection, and more.

"We already have 2D cameras which are used to provide lane-departure warningsthe camera looks out the front and tracks the dotted line. Cameras are also looking out and reading speed limit signs. Cameras can also look in your blind spot and alert you if a vehicle is about to overtake you," said Alexander. "A new application might be pre-crash sensing that calculates that a crash is imminent and pre-charged the brakes as well as fires the air bags more quickly than when waiting for the crash to happen."

One of the most promising approaches has been honed by Canesta Inc. over the last seven years since its founding in 1999. Canesta's CMOS image chip detects the distance to every object in a scene simultaneously, as opposed to ultrasonic that only senses distance to the nearest objects, by virtue of its time-of-flight calculations performed in hardware for every pixel on its imaging chip.

Canesta's CMOS imager chip simultaneously detects distance to every object in a scene.

Promising approach
"Canesta's approach looks promising, because it uses only a single camera and is based on CMOS technology which is critical to keeping costs low," said Alexander. "Plus its use of the time-of-flight method of calculating distance enables all types of automotive applications."

Time-of-flight refers to Canesta's use of an IR light source to illuminate scenes with invisible light, then measure the time it takes that light to fly from the emitter, which is beside the imager, out to the object, then back to the CMOS detector. By using hardware on the CMOS chip to calculate distance for every pixel in a scene, machine vision algorithms can easily group elements into objects, essentially perceiving rather than just sensing.

"We use a standard .18?m CMOS process that any foundry can provide," said Jim Spare, president and CEO at Canesta. "Also our SunShield technology enables us to perceive depth regardless of the lighting conditions rather than wait for an adaptive algorithm to kick in."

SunShield adds circuitry at every pixel, which senses the difference between the ambient light every 100?svirtually instantaneously&mdhas;compared to adaptive software algorithms, which can take milliseconds to adapt to changing light conditions.

Varying light conditions
"One of the things that distinguishes Canesta is its Sunshield technology that solves one of the hardest problems of using a camerathat is, it must be able to work in varying light situations to be safe enough to deploy in consumer automobiles," said Alexander. "I think Canesta is really on to something there, but they still have a ways to go to crack all these different automotive applications. I believe that it will be the 2008 and 2009 models in which we will see the first Canesta's 3D camera in consumer vehicles."

Canesta's first deployed automotive application will likely appear in new Hondas that use its 3D camera in the cockpit headliner to sense the size of a passenger and appropriately meter airbag deployment. Today this function is performed by pressure pads in the seats, but using a single CMOS camera may be just as cheap as pressure pads, while enabling other applications to share the camera, such as a "drowsy" alarm that can perceive the difference between blinking and when a driver's eyes are beginning to shut from lack of sleep.

Other companies are also experimenting with making CMOS image chips that detect depth on a pixel-by-pixel basis, notably International Electronics & Engineering S.A. Another company, Mobileye Inc., is sidestepping the 3D imagers, opting instead to use a normal 2D camera with their CMOS hardware accelerator chip that analyzes perspective to calculate distance.

- R. Colin Johnson
EE Times

Article Comments - CMOS imager chips enable true machin...
*? You can enter [0] more charecters.
*Verify code:


Visit Asia Webinars to learn about the latest in technology and get practical design tips.

Back to Top