Image sensor with parallel signal processing for motion detection

摘要

The paper describes the design of an unconventional biologically inspired image sensor. It contains numerous optical channels similar to facets in a natural compound eye. Each channel has several photodetectors with pre-amplifiers and a microcontroller with a multi-channel analog-to-digital converter. The signals coming from the photodetectors in each channel are amplified, converted into a digital form and processed by a microcontroller. All channels independently perform parallel image processing and image analysis. All microcontrollers are attached to a microcontroller network. They send data through this network only if useful signals are registered. These microcontrollers can be reprogrammed to perform various image processing operations including gradient search, spatial filtration, temporal filtration, signal correlation, neural network simulation and others. This design completely differs from the traditional image sensor architecture, which includes a mega-pixel focal plane array with a sequential signal read-out and a multi-core digital signal processor. The proposed architecture can be considered as a big set of the equal channels - the "smart groups'" of several pixels with read-out electronics and a digital microcontroller that can extract only the useful data and send it out. The working prototype of this image sensor has demonstrated ability to measure distribution of speed and direction of optical flow thought its field of view in the very short time. The advantages and possible applications of this sensor are also discussed.