Suitable For Robot Vision Algorithm Development And System Implementation

Machine vision is a relatively new and hot subject, which is expected to realize object tracking, recognition and measurement by using cameras and computer in stead of human beings. With the rapid development of computer science, image processing, pattern recognition, more and more attention has been paid to machine vision, which has resulted in many prominent achievements in some research fields. However, the large data and information load contained by image signal put much difficulty in real-time image processing, though they are definitely helpful to realize machine intelligence in the future. It is exactly one of the main bottle-necks in machine vision field. With regards to this difficulty, this paper is proposed to deeply explore the existing image processing algorithms and vision systems, analyze their practicability and practicability in robots, and hence develop a novel and more general vision system adapted to mobile robots.This paper is a summary of my work during my graduate study in ZJU. In the first chapter, some traditional algorithms and state of arts concerning image processing and robot vision are briefly introduced. Then our works are presented detailedly in the following chapters. Specifically, it consists of the following three parts: (1) Development of an improved object tracking algorithm based on image correlation; where we developed a novel tracking algorithm based on image. (2) Development of a localization algorithm applied to some special objects based on prior knowledge. In this paper, we used an automatic vehicle license plate localization system to demonstrate our adaptive localization algorithms regarding some special objects, where prior knowledge of these objects were taken into consideration, such as size, shape, color, etc. (3) A novel local invariant descriptor adapted to mobile robots. Here, a novel local invariant descript was constructed and then employed to describe objects so as to distinguish different objects. Both simulation and experimental results show that our algorithms have good tracking and recognition ability, and they are expected to have a super performance in real applications.