A Robust Method for Lane Detection under Adverse Weather and Illumination Conditions Using Convolutional Neural Network

Every year thousands of people lose their lives due to traffic accidents. Road accidents, especially the ones on the highways are the most fatal ones. Accidents not only cut peoples' lives short but also cause intense financial loss to the country. Many people who survive disastrous accidents are often left with critical injury or paralysis. Particularly in Bangladesh majority of the drivers are minimally educated. They do not have sufficient knowledge and tend to ignore the traffic rules often. As a result the roads are filled with careless drivers. Consequently most of the accidents on the highways and city roads occur due to the lack of awareness of the drivers. Additionally, many of the roads are poorly lit which makes it difficult to drive in unfavorable weather. An autonomous lane detection system can play an important role as a solution to the problem by assisting the driver in seeing the lane clearly. It can also generate warning to the driver in case of an unintentional or incorrect change in lane to avoid accidents. The lane detection method can be further developed to traffic sign and pedestrian detection and eventually a self-driving vehicle. In this study a robust lane detection method using deep learning has been proposed which can detect lanes in various weather and lighting situations. The proposed system has been compared to other baselines in related field demonstrates high accuracy and real time performance.