Adaptation of AI Road Surface Luminance Prediction model

Abstract

Brightness is a crucial factor in traffic safety operations. The photometric value commonly used to assess brightness is luminance (measured in cd/sq.m). Traditionally, acquiring this value requires expensive equipment and a long monitoring time. However, with advancements in artificial intelligence, it's now possible to predict luminance from pixels in a High Dynamic Range (HDR) image or a Low Dynamic Range (LDR) image, taken from either a professional camera (DHLR) or a smartphone camera.

The model was trained on a HDR dataset consisting of images from professional cameras with accurate luminance values. This allows the model to be used to recreate a luminance image from either an HDR image or an LDR image captured with a less expensive camera, reducing the need for expensive measurement equipment. The study's results demonstrate that the model can predict luminance from HDR images with promising visual results. Additionally, the luminance value can be combined with spatial information to identify areas with insufficient brightness or light uniformity.

This proposed method can measure luminance from road surfaces at night more efficiently, quickly, and inexpensively, making it beneficial for improving road safety.