An international team of scientists has introduced a miniature photomemristor that mimics the human eye's adaptation to bright and dim light. This innovation could benefit machine vision systems in robots, drones, and cameras, according to a study published in Nature Communications.
The device addresses the challenge faced by machine vision systems that lose accuracy during sudden changes in brightness. This is critical for drones and robots, which need to distinguish objects in dark areas while also detecting bright light sources like oncoming car headlights.
This development falls under the category of neuromorphic machine vision, where sensors not only capture images but also perform part of the signal processing. This approach aims to reduce the load on computational systems and speed up responses to changes in the frame.
Event-based cameras tackle a similar issue. Instead of capturing every frame in full, they register changes in brightness at individual pixels, resulting in low latency, high dynamic range, and reduced data volume. However, these systems require specialized algorithms and currently have their limitations.
The prototype is based on a miniature light-sensitive element measuring about 0.5 mm. The key component of the system is a photomemristor made from TiO2 and PEDOT:PSS. Its operation relies on the materials' response to humidity: in low light, the structure absorbs more water, increasing conductivity and light sensitivity. In bright light, moisture evaporates, reducing sensitivity.
In a demonstration setup, researchers used a 4 × 4 array of photomemristors and an artificial neural network. The system recognized letter patterns against backgrounds with varying brightness levels. According to the article, the accuracy was 91.3% under mixed lighting conditions, and the recognition process took 7.5 seconds.
In May, scientists introduced Qumus, an autonomous AI system for experiments with quantum materials.