High-resolution and whole-body hand contact dataset from grasping and manipulation based on 3D scanning method

Hand contact data, reflecting the intricate behaviors of human hands during object operation, exhibits significant potential for analyzing hand operation patterns, establishing guidelines for hand-related sensor and robot design, and predicting object properties. However, the translation from human hand operation patterns to practical insights for these potential applications is hindered by the constraints of low resolution and incomplete capture of the hand area within the contact data.Leveraging the non-contact and high-precision 3D scanning technology for surface capture, a high-resolution and whole-body hand contact dataset is constructed in this work. A total of 12 participants employed 52 kinds of hand gestures to operate 10 different objects. The contact area during each operation is whole-body painted on gloves, which are captured as the original hand contact data through a 3D scanner in the form of high-resolution point cloud. A coordinate conversion is applied to unify the coordinate of the original contact data and facilitate a clearer observation of the positional relationship of the data. Reliability validation on the obtained hand contact dataset is conducted, demonstrating an average similarity of 90%. Finally, hand movement classification with 95% precision is achieved through a neural network-based recognition model which is trained using the acquired hand contact dataset. We believe such data collection can be used as a stepping stone to develop machine learning techniques for grasping and manipulating with potential applications in hand posture recognition and hand movement prediction.