Inspired by Formula Student (FSAE), NTHU (National Tsing Hua University) Racing was established in 2015 and since then has placed in competition in categories such as Best Electric Vehicle, Best Electronic System Design, and Efficiency. Autonomous vehicles have increasingly become the focus of intense research interest which prompted students to take on a new challenge - the development of an autonomous racing car. The Imaging Source provided USB 3.0 industrial cameras to help the team reach the finish line.

Localization and Mapping
In order to test hardware and software designs, the students began with a scaled-down concept car (Fig. 1, below). The students' software approach would include multiple sensor types (i.e. sensor fusion) which would deliver the best mix of data for robust sensing and navigation. Vision sensors have long been integral to autonomous vehicles (AVs) for the visualization of objects in the environment such as signs and lane markings. But visual data can also be used to solve another fundamental challenge facing any autonomous vehicle: the ability to determine its own position in 3D space.

The team integrated a single-camera system using The Imaging Source's USB 3.0 DFK 33UX273 industrial camera. The camera's visual data could be analyzed for environmental information (e.g. yellow and blue pylons - Fig. 2, below) as well as for frame-to-frame visual tracking (monocular visual odometry). Using the Extended Kalman Filter (EKF) and FastSLAM-1.0 algorithms, the underlying structure and geometry of the environment was measured. The features were then aggregated in a visual map used to estimate the car's position with high precision (Simultaneous Localization and Mapping - SLAM).

Challenges and Breakthroughs
During testing, the team determined that the field of view (FoV) provided by a single camera was too narrow (Fig. 3, below) and was causing the system to miss some pylons. Drawing on the experience of other racing teams, the students decided to test adding a second camera to the system, increasing the FoV and capturing all environmental obstacles . The USB 3.0 camera connected to the computer quickly and efficiently, transferring the image information from the camera to the database in real time.