Autonomous driving solely dependent on camera input and classical computer vision methods

Autonomous driving is expected to be the next revolution in transportation. In recognition of this, the RDW each year hosts a self-driving challenge. Within the context of this contest, this thesis focuses on how autonomous driving can be achieved solely dependent on camera input and classical computer vision methods. The focus of the thesis is placed on lane detection, vanishing point based control, real time image stitching and optical flow odometry. Lane detection showed great results for multiple scenarios. Vanishing point based control worked really well as long as both road sidelines are visible. Camera stitching was unfortunately not achieved in real time due to poor environmental keypoints and unstable cameras, but only when manually calibrated on a video frame. Optical flow based odometry shows decent results but is likely out competed by IMU units and GPS.