Application of smart-pixel and microlens arrays to early vision

We are developing smart sensors that perform retina-like functions in the analog optical domain using optically cascaded smart-pixel arrays. By performing operations such as center-surround or oriented edge detection in a fully parallel fashion before the image is digitized and passed on to a computer, such smart sensors can greatly reduce the computational overhead associated with preprocessing. In order to implement the smart-pixel arrays, we have developed and demonstrated a device known as the monolithic optoelectronic transistor (MOET). In this paper we discuss the advantages of the proposed implementation, such as compactness, manufacturability, and broad spectral coverage, over approaches involving diffractive optical elements. We also discuss the fabrication and performance of the MOET and the mass transported lenslet arrays