| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程技術研究所 === 86 === Interests in studying the interfaces of object manipulation have continued to grow, especially for the application of immersive virtual environments. To achieve more reality in object manipulation, the glove-based input devices arecommonly chosen as the human-machine interfaces. In practice, the tracking devicesare also included to get the positions and orientations of the hands in the real world. Unfortunately, most of the hand-tracking gloves currently marketed are high prices, so they are not practical for widespread applications. In this paper, we present the development of a low-price data glove system using infra-red receivers/transmitters as the finger-bend measurement sensors. Not as convenient as the high-price ones, this data glove produces nonlinear outputs that must be calibrated before using it in a virtual environment. To make the glove easy for use, a four-stage calibration procedure together with the construction of the calibration device is developed. In the software calibration process, we devise a neural-network-based function approximator trained with a modified robust backpropagation (BP) algorithm which has the ability of eliminating the effect of noises in the training data. In order to speed up the training process, we propose a "tentative-and-refined" training method that is combined with a robust BP algorithm to constitute the modified one. Many successful experiments are made on a concrete data glove to verify the effectiveness of the proposed algorithms. So far, the experimental results of the calibration process with our method are very satisfactory.
|
|---|
