RGB Camera and LiDAR Fusion for Road Detection

Drivable road detection is one of the most fundamental problems for autonomous vehicles. Although good results can be obtained by using deep learning-based image segmentation methods on RGB camera images recently, the camera's dependence on ambient lighting increases the error rate in bad lighting conditions (such as dark environments, shadows, and reflections). Studies have shown that images obtained from cameras can produce better results when used with LiDARs by utilizing various fusion techniques. In this study, comparisons were made on U-Net model with early fusion by applying various fusion operations on RGB camera and LiDAR data based on the fusion methods in the literature. As a result of the study, sensor fusion approaches that give the best results in the field of driveable road detection were determined and inferences were made about how to improve success rates.