Efimia Panagiotaki

Master Thesis
Supervisors: Dr. Dengxin Dai

An Efficient Track Detection and Mapping System for Self-Driving Racing Cars

This work presents an efficient system for track detection and mapping for self-driving racing cars. The goal is to generate an accurate semantic map of the track and to provide accurate velocity and pose estimates. For the sensor system, we use two monocular cameras and an Inertial Navigation System (INS) in order to build an accurate and secluded visual SLAM system totally from scratch. The system is optimized for large scale applications and outdoors conditions. To detect cones' positions, we have implemented a cone detection algorithm which classifies SLAM tracked landmarks to "cones" and "background". The combination of cone detection, stereo SLAM and visual odometry builds a 3D map of cone landmarks. The system has proven to be robust to fast angular rates, and successful tracking was achieved for more than 180°/s. The method has been integrated into a self-driving system for racing car Flüela. Flüela is a lightweight, 4 wheel drive electric race car developed by Akademischer Motorsportverein Zürich (AMZ) Driverless. In August 2017, AMZ participated in the Formula Student Driverless Competition in Hockenheimring, Germany, and has won the 1st place overall, among 15 other driverless teams.