BALTIMORE Harnessing neural-network technology gleaned from the biology of the human eye, a Johns Hopkins University researcher has demonstrated a "smart eye" chip that combines image sensing with filtering and object-tracking capabilities. The brainchild of Ralph Etienne-Cummings, the circuit is based on a hybrid analog-digital CMOS technology and is touted as being much faster than distributed-component systems that use separate chips such as an image sensor, microcontroller and non-volatile RAM to control image-processing software.
Ralph Etienne-Cummings of Johns
Hopkins tests robot's artificial eye.
"I was still in school when Carver Mead started the idea of neuromorphic engineering basing your electronic designs on biological blueprints," Etienne-Cummings said. The new chip, he said, "mimics the eye by surrounding a high-definition central region of pixels that are very sensitive to movement with a low-resolution peripheral-vision area that tracks the location of objects so as to keep the central region centered on them."
Etienne-Cummings took advantage of the parasitic bipolar transistors inherent in CMOS chips, fashioning them into surface arrays of photosensitive pixels. The analog interconnecting matrix for these bipolar transistors implements motion detection in the central region, deriving speed calculations from it. The periphery analog interconnection matrix figures out the location of the object by bumping against the edge of its central region, deriving a heading calculation from it.
That, together with the speed calculation, is used to lock onto and track moving objects. All operations are executed directly by the sensor chip itself, without the use of a microprocessor.
"The speed with which the chip can track objects is orders of magnitude faster than you can ordinarily achieve. I hope to someday have helped create a technology that will enable doctors to track movement of a beating heart so that blocked cardiac arteries can be cleared without having to stop the heart first, as doctors must do today," Etienne-Cummings said.
The technology's capabilities are being demonstrated by mounting the single-chip smart eye on a small mobile robot, enabling the robot to follow a line around the floor while avoiding obstacles in its path. In this application, the motion of the robot makes even stationary objects move within its field of vision. Etienne-Cummings has successfully demonstrated that the imaging system can track objects moving within its field of vision no matter what the source of movement, whether it's the object moving or the robot moving.
"This system is much smaller and uses much less power for all kinds of new mobile applications in microrobots, autonomous flying machines and extraterrestrial rovers," Etienne-Cummings said.
He listed possible applications in autonomous navigation, medical systems, pick-and-place parts manufacturing or videoconferencing systems that "lock on" to a speaker as he or she moves around the room.
The mobile-robot demo used two chips, one for each "eye," in an arrangement that favored avoiding obstacles over following a line. When avoiding an object caused it to lose sight of the line, the robot remembered which way to steer to get back to the line.