Abstract
The objective of Selective Change Driven (SCD) Vision is to capture and process those scene pixels that have the greatest impact in the motion estimation task. The implemented SCD Vision sensor delivers the pixels ordered according to the illu- mination change undergone by each pixel, from the last time each pixel was read-out. This ordering strategy is especially interesting for motion detection algorithms, since it allows for a reduction in data bandwidth requirements without decreasing accuracy. The speed of the obtained pixel flow allows movement detection and tracking at a speed several orders of magnitude higher than conventional vision systems. To ac complish these objectives, the sensor is based on a well-known logarithm continuous time cell for light capture, and a new high-speed several thousand input Winner-Take-All (WTA) circuit with single winner selection. This WTA circuit differentiates this sensor from others, since it allows for a constant synchronous pixel rate, self-adaptation to scene conditions, the operation being controlled by the processing hard- ware, the complete use of the available processing and bandwid th resources, and illumination and address event delivery at the same time resolution of 2 µs. A 64 × 64 SCD Vision sensor is presented w h i c hh a sb e e ni m p l e m e n t e di ns t a n d a r d1 8 0n mC M O S .