Multi Sensor based Approach for Road Region Extraction for Autonomous Vehicles

Automatic road region extraction is an important component of intelligent and driverless vehicles towards providing smooth navigation. This paper presents a novel method of extracting road region using LiDAR and camera in coherence. The ground region is extracted from LiDAR while camera also gives input on the current scene in parallel. The data obtained from LiDAR and camera is mapped using trigonometric equations. A patch from the camera data is extracted and pixel values of the patch are sent to Gaussian Mixture Model-Expectation Maximization (GMM-EM) algorithm for training at periodic intervals. Based on this training, a drivable region is identified in every frame of the camera input. Thus, obtained drivable road region is further processed and is used in decision-making during navigation. The proposed system works efficiently on different kinds of roads with different lighting conditions and gives a good estimation of the drivable road region in presence of objects or obstacles.