There have been many research efforts to develop an intelligent robot system. Conventional robotic system has been performing their task which was taught by a human operator previously. However, more complicated working environment requires an advanced robotic system which can adapt itself to environmental changes. Vision can provide many useful information for the robot. In most cases of Eye-in-Hand robot, a single camera is fixed at the robot end-effector, which causes complex image processing.
One of the most important difficulties is the extraction of the real-time 3D information and specially an real-time estimation of depth or range information. In this thesis, depth estimation in the image from hand-held camera is studied and a fuzzy controller which can use uncertain image features to the robot control effectively is also presented.
To estimate the depth, two different clues of 3-D information in the images which can be obtained by a single camera are described. One of many practical problem in the Eye-in-Hand robot is blurring of the image. Because the distance between the camera and target is varied according to the robot control, we cannot get clear image for the whole robot control. Except an in-focusing distance region, the images taken from the lens have blurred. However, this blur provides a clue to estimate the depth. And it is useful depth estimation approach in Eye-in-hand since depth-from-focus requires only one image and several lens parameters.
However, the conventional blur measurement methods may suffer from noise contamination or computational complexity which makes their real-time implementation difficult. And depth-from-focus often gives large errors due to the distortion of the blur. In this thesis, a simple and robust blur measurement is proposed and a fuzzy-rule based depth determination is also studied.
As a real-implementation, an advanced image processing hardware is developed with HAU(Hardwarized Algorithm Unit) which can implement various image processing algorithms into hardware case by case. As another depth clue, depth-from-motion which can handle the moving object is also considered. It can be used for a long distance. To provide depth information for Eye-in-Hand robot, those two depth estimation approaches are integrated.
In spite of the many researches, 3D information from image still contains some errors depending on the real depth and environment. To control the robot by using visual information, the controller has to be designed to deal with the uncertainty of the information. To meet these requirements, a self-organized fuzzy controller has been proposed.