Monocular vision, optical blur and visual looming as a basisfor mobile robot obstacle avoidance

TLDR: It was found that obstacle approach could reliably be computed through this method, but its success depended on the camera lens properties, and it was intended that this system be relatively simple in hardware and software implementation.

Abstract: Avoidance of obstacles is a fundamental problem for mobile robots. In recent times, a rapid increase in computing power combined with an equally steep decrease in size, electrical requirements and cost in computers has led to the increase in the use of cameras and vision as a realistic means of achieving real time avoidance. In this paper, we investigate a means by which a machine vision system could utilise optical blur as an avoidance indicator. The methods used are intended for monocular systems and employ the blur recovery methods of Hu and de Haan to find optical blur and optical blur and the looming method described by Raviv and Joarder and Sahin and Gaudiano to relate this to object approach. It was intended that this system be relatively simple in hardware and software implementation. To verify the success of the design, we conducted tests in a controlled environment. It was found that obstacle approach could reliably be computed through this method, but its success depended on the camera lens properties.