UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

CMOS image sensor design with programmable spatial-temporal exposure for machine vision and computational imaging applications Luo, Yi

Abstract

Compressive sensing, as one of computational imaging techniques, employs exposure encoding of cameras. Currently, as coded exposure is not supported monolithically on image sensors, computational cameras rely on discrete optical modulators to implement compressive sensing. In this thesis, we propose image sensor designs that are capable of per-frame spatial-temporal exposure encoding. We propose merging exposure-programmable pixels, which consist of charge modulators and exposure-code memory, into the imager design. Through pixel-wise exposure manipulation in every frame of image capture, compressive sensing and its related imaging applications are extended to the sensor node with significant benefits of high optical throughput, improved power efficiency, and compact footprint. In the design of exposure-programmable pixels, four types of pixel architectures are proposed. The capacitive-transimpedance-amplifier-based pixels are advantageous in sensitivity and charge-transfer speed, while the other two which use active-pixel-sensor-based structures offer a more compact size and circuit simplicity. To exploit the full potential of proposed pixel designs, the image sensor architecture is correspondingly modified as compared to the conventional image sensor designs. To evaluate the feasibility and performance of the proposed designs, two prototype image sensors are fabricated in a CMOS process. From experimental results, both conventional non-intermittent exposure and per-frame spatial-temporal coded exposure are verified. In demonstration of on-chip compressive sensing applications, two examples from high-speed imaging and compressive focal-stack depth sensing are presented. By performing compressive sensing at the sensor level, the CMOS image sensor designs introduced in this work further pave the way to on-chip computational imaging and facilitate implementation of many emerging applications in the machine vision paradigm.

Item Media

Item Citations and Data

Rights

Attribution-NonCommercial-NoDerivatives 4.0 International