Qiushi Fu

This paper proposes a framework for tracking both human hand kinematics and object contour during grasping task. The framework is based on modeling the object as point cloud and the use of marker-based tracking. We introduce how to estimate contact sites on both the hand and object, hand enclosing space, and graspable features from recorded data. Two experiments were performed to 1) verify the accuracy of contact site estimation (less than 5 mm), and 2) validate the feature extraction. Our approach can provide significant insight into how humans plan grasping and manipulation based on object recognition.

This paper describes the general procedure, model construction, and experimental results of tracking whole hand kinematics using extended Kalman filter (EKF) based on data recorded from active surface markers. We used a hand model with 29 degrees of freedom that consists of hand global posture, wrist, and digits. The marker protocol had 4 markers on the distal forearm and 20 markers on the dorsal surface of the joints of the digits. To reduce computational load, we divided the state space into four sub-spaces, each of which were estimated with an EKF in a specific order. We tested our framework and found reasonably accurate results (2-4 mm tip position error) when sampling tip to tip pinch at 120 Hz.