A VISION-BASED 3D MOUSE

Abstract

A novel scheme using a mirror and a single camera to restore the 3D position of a finger tip is introduced. In such a scheme, an algorithm which integrates Bayesian estimation into the traditional chain code generation algorithm is developed to improve the contour detection on a hexagonal grid on the fixed but complex background. In our algorithm, we do not use prior shape knowledge as this would reduce the usability and robustness of a practical 3D mouse. By integrating the smoothness assumption, the chain code generation algorithm produces the smoother outputs in high speed. Post-processing which utilizes the constraints provided by the system setting and signal filters produces smooth 3D coordinates stream. Finally a demon is built and the performance of the 3D mouse are thoroughly analyzed.