Fingernail imaging is a method of sensing finger force using the color patterns on the nail and surrounding skin. These patterns form as the underlying tissue is compressed and blood pools in the surrounding vessels. Photos of the finger and surrounding skin may be correlated to the magnitude and direction of force on the fingerpad.;An automated calibration routine is developed to improve the data-collection process. This includes a novel hybrid force/position controller that manages the interaction between the fingerpad and a flat surface, implemented on a Magnetic Levitation Haptic Device. The kinematic and dynamics parameters of the system are characterized in order to appropriately design a nonlinear compensator. The controller settles within 0.13 s with less than 30% overshoot.;A new registration technique, based on Active Appearance Models, is presented. Since this method accounts for the variation inherent in the finger, it reduces registration and force prediction errors while removing the need to tune registration parameters or reject unregistered images. Modifications to the standard model are also investigated. The number of landmark points is reduced to 25 points with no loss of accuracy, while the use of the green channel is found to have no significant effect on either registration or force prediction accuracy.;Several force prediction models are characterized, and the EigenNail Magnitude Model, a Principal Component Regression model on the gray-level intensity, is shown to fit the data most accurately. The mean force prediction error using this prediction and modeling method is 0.55 N. White LEDs and green LEDs are shown to have no statistically significant effect on registration or force prediction. Finally, two different calibration grid designs are compared and found to have no significant effect.;Together, these improvements prepare the way for fingernail imaging to be used in less controlled situations. With a wider range of calibration data and a more robust registration method, a larger range of force data may be predicted. Potential applications for this technology include human-computer interaction and measuring finger interaction forces during grasping experiments.
展开▼
