Trading sharpness with energy consumption in a lens autofocus application
Published in Workshop on Approximate Computing, Embedded Systems Week, 2016
Recommended citation: Molnos, A. (2016). "Trading sharpness with energy consumption in a lens autofocus application" Workshop on Approximate Computing, Embedded Systems Week.
Anca Molnos, Yves Durand and Nicolas Gonthier
Abstract
Approximate computing is an emerging paradigm in which the accuracy of computation results can be traded against, e.g., savings in energy, improvement in performance. In this extented abstract we investigate, by means of an example, the applicability of approximate operators (additions, in our case) on an adaptive feedback control loop. Our research vehicle is a autofocus controller that sets the focal distance of a lens such that a defined region of interest (ROI) in the output image is sharp. We study the energy consumption and the ROI sharpness error for various operation approximation degree. The results are encouraging indicating that for a 30% reduction on the energy consumed in the addition operations the degradation in the sharpness of only 2%. A 40% reduction on the energy consumed corresponds to a less than 10% degradation.
Recommended citation: Molnos A., Durand Y., Gonthier N. (2016) “Trading sharpness with energy consumption in a lens autofocus application”. Workshop on Approximate Computing, Embedded Systems Week.