Pixels for focal-plane scale space generation and for high dynamic range imaging

Abstract

Focal-plane processing allows for parallel processing throughout the entire pixel matrix, which can help increasing the speed of vision systems. The fabrication of circuits inside the pixel matrix increases the pixel pitch and reduces the fill factor, which leads to reduced image quality. To take advantage of the focal-plane processing capabilities and minimize image quality reduction, we first consider the inclusion of only two extra transistors in the pixel, allowing for scale space generation at the focal plane. We assess the conditions in which the proposed circuitry is advantageous. We perform a time and energy analysis of this approach in comparison to a digital solution. Considering that a SAR ADC per column is used and the clock frequency is equal to 5.6 MHz, the proposed analysis show that the focal-plane approach is 26 times faster if the digital solution uses 10 processing elements, and 49 times more energy-efficient. Another way of taking advantage of the focal-plane signal processing capability is by using focal-plane processing for increasing image quality itself, such as in the case of high dynamic range imaging pixels. This work also presents the design and study of a pixel that captures high dynamic range images by sensing the matrix average luminance, and then adjusting the integration time of each pixel according to the global average and to the local value of the pixel. This pixel was implemented considering small structural variations, such as different photodiode sizes for global average luminance measurement. Schematic and post-layout simulations were performed with the implemented pixel using an input image of 76 dB, presenting results with details in both dark and bright image areas.