Ambient Temperature Estimation of Mountain Freeways Based on Roadside Camera Images

Abstract

Accurate estimation of the ambient temperature of mountain freeways enables freeway management agencies to provide weather-related information to drivers. This article proposed an image-based data-driven method, namely the visual temperature estimation network (VTENet), to estimate freeway ambient temperature based on images captured by roadside cameras. The VTENet had a convolutional neural network (CNN) architecture to extract temperature-related image features, and two extra networks to capture space-time information on data collection and time-series image features. The VTENet was trained and tested based on a self-established dataset collected at a mountain freeway. The results showed that the VTENet can estimate freeway ambient temperature with high accuracy. The model gives a more accurate temperature estimation with data collected from 10 to 11 A.M. and 2 to 3 P.M. than other periods. It also performed better using four-day or five-day sequence images than other data inputs.