Vision systems-an overview

Vision is one of the most difficult core technologies, but is essential for robotic automation. Human vision is so natural that it is hard to appreciate the complexities of creating it artificially, but vision is the only thing holding back the development of truly autonomous machine systems. Machine vision systems are designed to eliminate or reduce human observation and error, such as NASA's Mars explorers being deployed in the late 1990's or Meta Machines Ltd.'s vision system that can withstand the high temperatures encountered during high-current welding. Through research into finding better algorithms and more sensitive cameras, the number of potential applications will increase due to the decrease in price. These applications fall into three categories: measurement, inspection and guidance. Measurement systems find the dimensions of an object through digitization and manipulation of the object's picture. Inspection systems determine whether an object matches a preset description. Guidance systems cause a machine to perform certain actions based on what it "sees". Vision research is difficult because the reasoning the human brain applies to identify objects cannot be easily transferred to a computer. Computer vision work traditionally has been based on statistical and heuristics, or trial and error methods to analyze image data. Common methods, such as edge detection, give unreliable information about the scene being viewed.