Lane detection to estimate road curvature and vehicle position for an assisted driving vehicle

Over the last ten years fuel economy (FE) has been improved thanks to the incorporation of a variety of ADAS sensors in hybrid vehicles. ADAS sensors make it possible to estimate road information such as the road slope and the road curvature. The aim of this thesis is the implementation of an advance lane detection algorithm that is able to compute the radius of the road curvature and the vehicle position with respect to the lanes’ centre. Two ADAS sensors are fundamental for this work: lidar and camera. The starting point is the implementation of a lane detection algorithm in 3D point cloud using a lidar sensor. The proposed Matlab algorithm allows to detect the lane points using the sliding window technique. On this basis the polynomial curve is estimated. This process allows to understand the algorithm of a simple lane detection. The second step consists in an advanced lane detection using a single camera sensor. The input images are taken from KITTI road dataset. Gradient and HLS thresholding are the techniques used to identify the lane line in binary images previously obtained. The lanes are visualized thanks to the sliding window search technique that estimates the lanes’ colour. Afterwards a terrain path with a series of curvatures is created, each with a different radius. A driving scenario is recorded on the terrain path using a simulation and validation software called SCANeR™ studio. The resulting recording is then given as input to the python code which processes an output video displaying the left and right lane detected with the respective radius of curvature. The final goal is to compare the measure of each curvature radius, that is part of the terrain path, with the measure of each radius estimated by the algorithm written in python in order to verify if they are equal.