Hand-Eye: A Vision-Based Approach to Data Glove Calibration

We describe a new method for data glove calibration that uses computer vision techniques to create a filter for individual customization of input obtained from the glove. Our major observation is that it is possible to create a linear correlation between the hand posture reported by the data glove and the observed posture of the hand itself. We use a feature-based computer vision system independently to extract information about hand posture from video images of a human hand that is using the data glove. We simultaneously collect the glove data for the same posture. Linear regression is used on the combined sets of reported data to establish a filter that customizes the data reported from the glove for an individual user. The filtered glove data is mapped onto a computer-generated image of the hand’s skeletal structure. We show by comparison that the computer-generated hand image exhibits a posture quite similar to that of the actual hand.