| dc.description.abstract | Cars, particularly manually-driven cars, are one of the most commonly used modes of transportation today. However, millions of people are either killed or left with disabilities annually due to road traffic accidents caused by human error or sensor failures. Despite that, a lot of people seem reluctant to look into alternatives to manually driven vehicle transportation. This is understandable as driving cars has been the trustworthy mode of transportation for many years, and it is widely used in everyday life around the world. However, technological advances in the fields of machine learning and cyber-physical systems contributed to the emergence of nearly or fully autonomous vehicles, or driverless cars, as a true viable alternative for the current human-controlled driving mode. The technology still has a long way to go, mainly because the advances in vision and depth measurement sensors such as LIDARs can not achieve the levels of safety needed to make fully autonomous cars. Progress on this front is being made every day, and it seems inevitable that they will be readily available in the near future. Our team aims to further investigate the application of Computer Vision and sensor fusion to achieve independent self-driving without external guides. To accomplish this, we combine a depth camera with a LiDAR to provide better coverage of the surroundings and allow more accurate detection and thus accurate avoidance of obstacles. We are mounting the vision system on a model driverless car and using the vision data to guide the car control system. A computer vision algorithm will be run by the NVIDIA Jetson Nano to determine what course of action the car should take. The final prototype should be capable of driving at a reasonable speed without colliding with any objects and making decisions such as braking or turning when necessary. | |
