Weather Classification on Roads for Drivers Assistance using Deep Transferred Features

Extreme weather conditions such as heavy rain, fog and scorching heat of sun increase the risk factor of road accidents and traffic congestion. This impels valuable lives and property into tremendous dangers and causes dilatory. Therefore, automatic weather detection on roads for the assistance of drivers becomes very crucial. Weather detection and forecasting is usually figured out by using temperature, humidity and wind sensors. In this paper, we introduce a methodology which exploits visual data to detect the weather condition. We propose a weather detection system based on state-of-the-art deep learning techniques by transferring the learned weights of pre-trained inception v3 model to our problem. Proposed system can be utilized to generate alerts for upcoming vehicle's drivers to change their driving behavior according to the weather condition. The system can also be helpful for Intelligent Transportation System (ITS) authorities so that they can provide road safety to the citizens with efficient way and improve the current ITS. The system is trained using self-generated dataset. Proposed system achieved an accuracy of 98% on unseen dataset, which is very high as compare to prior models.