Development of a real-time vision based absolute orientation sensor.

摘要

In many nonlinear systems such as multiple degree of freedom wrist-like actuators, absolute measurement of the orientation is essential to control such devices. In this thesis a real-time vision based absolute orientation sensor has been developed to measure the absolute orientation of a spherical body for use in real-time control with specific application to the rotor of the variable reluctance (VR) spherical motor. This sensor has the significant advantage over those using multiple single-axis encoders by providing a non-contact means of absolute orientation measurement, which has the potential to simplify the overall system dynamics and give improved performance. The specific objectives of this thesis are the development of a technique for recovering the absolute orientation of a spherical body through analysis of a specially designed grid pattern on the body, and a means of modeling such vision based measurement systems. These objectives have been accomplished by the following tasks. First, a new optimized grid pattern design suitable for direct application to the rotor of the VR spherical motor has been developed along with more practical techniques for transferring such a pattern onto a spherical surface. To aid in this design and other important aspects related to the evaluation of the vision based absolute orientation sensor, an OpenGL based grid pattern image modeling environment has been developed. Efficient image processing techniques suitable for real-time machine vision applications have also been developed for analysis of the grid pattern images. These image processing results have also been used in a mathematically sound procedure for high-resolution recovery of the absolute orientation that utilizes camera calibration and a closed form orientation recovery technique well suited for real-time applications. Finally, experimental tests and an OpenGL based grid pattern modeling environment have been used to verify the capability of the absolute orientation measurement system to make high-resolution measurements. It is expected that this research will extend the use of vision based measurement systems to real-time control applications that require simultaneous multiple degree-of-freedom absolute measurements.