AI for Lighting Tunnels

An abrupt change in lighting is a tremendous safety hazard for drivers in short tunnels (80 ft – 410 ft). While artificial lighting is required in longer tunnels regardless of the outside lighting conditions, road safety can be significantly improved in short tunnels by adjusting the artificial lighting accordingly with the lighting conditions outside. In this research, we develop an intelligent lighting system to adjust the artificial lighting inside the tunnel based on the instantaneous lighting condition outside. To this end, we use the data collected from 13 tunnels throughout Georgia to develop a power regression model, which can then be used to estimate the illuminance distribution in short tunnels. The proposed model is then validated using the results of a 3D finite-element simulation model. We use a machine learning approach to re-estimate the parameters of the illuminance prediction model based on the tunnel features, for example: tunnel type, geometry, and length. Using the portal/sky illuminance data from an optical sensor outside the tunnel, we can accurately predict the illuminance distribution and dynamically adjust the artificial lighting inside short tunnels. In addition to improvement in road safety, the proposed intelligent lighting system significantly reduces the consumption of electricity for lighting short tunnels, which justifies switching to solar batteries as a reliable electricity source in practice. Considering the high cost of manual assessment, wire installation, and electricity consumption off the grid, the proposed intelligent lighting system can save the Georgia Department of Transportation (GDOT) a significant amount of time and energy every year. The artificial intelligence model developed in this project will have broader applications in lighting short tunnels to enhance the safety of roads on the national scale. Keywords: short tunnel lighting; finite-element simulation, power regression; machine learning; optical sensors; artificial intelligenc