An investigation of the use of vision systems for the robotic control of automated vehicles

Abstract

Humans’ usage of the motor vehicle for transport and freight is ever increasing. It seems that the risks and level of accident rates associated with these traffic systems can only be lessened by increasingly complex systems, which aid the human driver with the task or take some of the task out of the human driver’s control. However, any semi or fully autonomous vehicle must conform to the type of roads currently in existence. Those roads are designed to suit a human sensory system, mainly the vision sense. Therefore, it would seem that computer vision type systems would be the largest contributor to autonomous vehicles in the short to medium term future. This work investigates how vision systems have already been used in one type of autonomous vehicle task, namely “lane detection and following”. It also implements algorithms that will accomplish this task from road image capture, through detection of lane markings to trajectory planning and steering controls required to traverse the planned trajectory. Each part of this overall algorithm is based on existing algorithms that are discussed in the literature review section. These algorithms are however, implemented or used in novel ways in this project. Then based on the results from running each section of this algorithm recommendations are made regarding the current usefulness of these methods and how each could be expanded upon and improved to be used as a viable solution in the future. Also given the dangers of testing new algorithms on the open highway, this work investigates the design of a model test system that could be used to test algorithms in a safe and compact controlled environment. Some small parts of this model test system are implemented in this work.