Lane keeping in roadwork conditions for a scaled autonomous car

Driving automation technologies have been gradually penetrating the automotive market. The road infrastructure may be temporarily modified by construction sites, requiring suitable adjustments for the vehicle driving task. This thesis aims to develop the lane keeping task in a simulated diversion scenario induced by roadworks, exploiting vision-based lane detection techniques. The vehicle engaged for the simulation is a 1/10 scaled self-driving car, provided by Bosch for the Bosch Future Mobility Challenge (BFMC). Two specific scenarios have been reproduced: lane change and lane narrowing. Color segmentation, performed by using OpenCV C++ libraries, has been crucial for prioritizing yellow lane markings over white standard ones, and for detecting predefined obstacles. Analysis have been conducted by evaluating the vehicle speed, steering angle and position with respect to the lateral lane line. The diversion has been correctly taken in most of the experiments for both lane narrowing and lane change. The presence of a traffic sign indicating the diversion path, and of red barriers blocking the closed lane, improved the lane change behavior. Partial success has been obtained for obstacle detection when the red barrier is positioned across the lane, forcing the vehicle to stop. Further refinement through additional testing should enhance driving performance, while sensor fusion could improve object detection. Future research should comprise the integration of additional traffic signs and traffic cones to make the roadwork environment more realistic.