Bioinspired in-sensor spectral adaptation for perceiving spectrally distinctive features

Abstract

In challenging lighting conditions, machine vision often yields low-quality results. In situations where particular spectral signatures carry critical information, adapting the spectral sensitivity of visions systems to match the predominant spectra of the surrounding environment can improve light capture and image quality. Here we report spectra-adapted vision sensors based on arrays of back-to-back photodiodes. The spectral sensitivity of these bioinspired sensors can be tuned to match either the broadband visible spectrum or a narrow band within the near-infrared spectrum by applying different bias voltages. The process of spectral adaptation takes tens of microseconds, which is comparable with the frame rate (around 100 kHz) of state-of-the-art high-speed cameras. The spectral adaptation increases the Weber contrast of the scene by over ten times, resulting in increased recognition accuracy (from 33% to 90%) of features when exposed to intense visible-light glare.