Weather-informed vision enhancement for autonomous vehicles in adverse conditions

Abstract

Delivering Advanced Driver Assistance System functionalities depends on acquiring high-resolution image data from vehicles. Adverse weather and nighttime degrade image quality, impacting object detection accuracy. This paper addresses this issue by proposing a novel solution using the vehicle's Global Positioning System location and timestamp to query weather via a weather application programming interface. By obtaining weather details at the time and location of data collection, image quality is enhanced through pre-processing tailored to specific weather conditions. Using the Detection in Adverse Weather Nature dataset, the method improves You Only Look Once version 8 software detection accuracy by up to 15% compared to baseline performance across various weather conditions, enhancing Advanced Driver Assistance System reliability.