Hand motion expertise analysis using dynamic hierarchical activity modeling and isomap

Several domains such as sports, surgery, dance etc. are characterized by a significant influence of expertise of the performer on the motion pattern and style. The retrieval of expertise level of the performer through automated motion analysis is the subject of this paper. We employ a novel neuropsychologically inspired algorithm that employs a dynamic hierarchical layered structure to represent the human anatomy, and low-level parameters to characterize motion in the layers of this hierarchy which correspond to different segments of the human body. This characterization is representative of the expertise of the performer of the motion. These motion profiles are then compared using dynamic time warping to generate a similarity matrix. We employ isomap to reduce dimensionality and the cluster the data into different expertise classes. This algorithm was tested on a library of surgical movements that contained 3D hand motion data of common surgical laparoscopic procedures. Linearly separable clusters were obtained between novice, intermediate and expert performances. Test sequences were projected into manifold spaces. A recognition percentage of 98.56% was obtained for classifying the test sequences into correct expertise clusters.