机器视觉维修系统英文文献和中文翻译(2)

菜单

detecting objects of interest that may be moving at • Systems which detect objects of interest only while times and stationary at other times. they are moving. Once objects of interest stop, they

This research goes beyond merely detecting the become invisible to the motion detection scheme. presence of an object. We also connect the detection • Systems which function properly only when the module to other important sensory components of a camera is either stationary or moving, but not vision-based robot control scheme. Particularly both. To the contrary, our system generally significant is the ability to find landmarks on objects operates under static camera conditions, but also of interest and to know about the projected shape of allows the freedom of visually servoing an eye-in- objects. In addition, tracking techniques are used to hand system based upon target location.

monitor objects without human intervention. Our • Systems which cope with a single moving target, solution to the tracking problem follows the even though several application areas involve controlled active vision framework,' which avoids a images with several targets.

heavy reliance on a priori information through the • Systems which assume that a moving object is a use of optical flow. Optical flow is induced by any rigid body. Further assumptions may include a combination of camera or object motion. specific pattern of motion for the object of interest.

The primary contribution of this research is a

complete software and hardware implementation of our detection framework. In the process of con- structing this system, we have selected and modified computer vision techniques that are appropriate to the visual detection problem. Many of the techniques used by our framework (e.g. frame-differencing) have also appeared in similar forms in existing research, which is a demonstration of their usefulness. Our solution to the detection problem is innovative in the way in which it has uniquely combined these techniques into a general framework that can be directly applied to real-world situations. We have made modifications to these techniques where necessary, and we have also incorporated our own ideas where the existing literature was lacking (e.g. dynamic segmentation domains). Finally, we have customized the theory to the specihc industrial needs of robotics. This has demonstrated that our frame- work can provide precisely the type of information required to manage effectively a situation requiring visual detection. Results from experimentation with our visual servoing systems show the potential of our approach under general conditions.