The skeleton of the human hand consists of 27 bones, including 8 short carpel bones. Current state-of-the-art in human hand modeling only supports an approximate version of human hand because of the difficulty in acquiring a "stable" (without hand movements) data of the internal hand anatomy, as well in segmentation algorithms for identifying bones in the captured anatomical images. Computed Tomography (CT) scans irradiate the hand with harmful radiation and do not have the contrast to show any anatomy other than bones. Magnetic Resonance Imaging (MRI) scanners can provide internal anatomy, but the acquisition is difficult because the hand must be kept still inside the MRI scanner and because a quality scanning session is long (about 10 minutes). The goal of this project is to greatly improve the modeling, simulation and animation of human hands. Such models are useful in many fields. In computer graphics, virtual reality, telecommunication and film, they enable better, more believable virtual hands, enhancing the immersive experience. Accurate hand models can be used to design tools, equipment and everyday objects that must be manipulated by hands. In healthcare, they improve the design of medical devices that come in contact with hands. In the apparel industry, they enable one to design better gloves. These computer models can also improve the design of robotic hands for medical prosthetics, enabling the artificial hands with artificial bones and muscles to move and deform like their real biological counterparts.
The project will develop a stable method to acquire hand internal anatomy (bones, muscles, fat) in multiple hand poses using MRI scanners, by manufacturing ergonomic rigid molds that hold the hand in a fixed and known pose during the scan. Real bones do not simply rotate around some center of rotation at the end of another bone, but instead undergo complex rigid body motion relative to their parent bones. Using the acquired bone rigid body motion, the research team will develop new methods to model this complex rigid motion, using novel data-driven and model-based techniques. Finite Element Methods (FEM) based simulations are then applied to combine acquired pose-varying muscle and fat/skin shapes. These advances enable realistic modeling of detailed surface appearance of hands that matches ground truth surface scans and that generalizes to arbitrary hand poses. The research team will then build a computer model for how the bones and muscles of the human hand move when the hand is articulated. Given the three-dimensional surface scans of the hand in a few poses, the research team also will use computer simulation to "fill-in" occlusions in scanned poses (occurring in a closed hand, fist pose and similar). This procedure will create a computer model of both the internal hand anatomy and the external hand appearance.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
