Abstract— Many studies of human postural control use data from video-captured discrete marker locations to analyze via complex inverse kinematic reconstruction the postural respon- ses to a perturbation. We propose here that Principal Compo- nent Analysis of this marker data provides a simpler way to get an overview of postural perturbation responses. Using short (1, 4, and 16 mm) anterior platform step translations that are on the order of a young adult’s normal sway path length, we find that the low order eigenmodes (which we call eigenposes) of the time-series marker data correspond dominantly to a simple anterior-posterior pendular motion about the ankle, and secondarily (and with less energy) to hip flexion and extension. A third much weaker mode is occasionally seen that is repre- sented by knee flexion. I. INTRODUCTION ntegration of motion capture devices into postural control experiments is essential for analysis of postural control dynamics because body configuration must be known. Using markers at key motion joints, we are able to accurately de- scribe body configurations during testing of postural control response using SLIPFALLS-STEPm, a controlled sliding platform test setup with an associated motion analysis sys- tem that tracks the motion of the markers in three dimen- sional space using a multi-camera system [1,2]. However, even if only a small number (~ 20) of markers are tracked in 3-D Euclidean space, the dynamic marker location data results in a 60-dimension time series. This high-dimensional data representation does not support analysis of postural control dynamics, except where complex inverse kinematics is employed with corresponding assumptions. These inverse kinematic calculations involved in postural stability and gait analyses require measurement of all appro- priate limb segments and joint angles by intensive inverse computation of marker positions. These calculations, coup- led with the force reactions of the feet with the platform on which a subject stands and anthropomorphic assumptions about the weight distributions in various body segments, are designed to yield joint torques and moments about each axis of rotation, and hence to discern reactions to perturbations. Manuscript received February 4, 2010. Research funds received from NIH R01 AG26553, VA Rehabilitation R&D #E2143PC, and a VA Senior Rehab Research Career Scientist Award. Dr. Robinson with the VA Med Center Research Service, Syracuse, NY; and the Center for Rehab. Engineering, Science and Technology, Clarkson Univ., Potsdam, NY.13699-5730 USA <c.robinson@ieee.org> Drs. Skufca and Bollt are with Clarkson Univ’s Math and Computer Science Dept. <(jdskufca, bolltem)@clarkson.edu>. Mr. Pilkar is a PhD student in the Electrical and Computer Engineering Dept. at Clarkson University <pilkarrb@clarkson.edu>. The most used measure of reaction force is Center of Pressure or CoP, which is the vertical projection of the body’s center of mass (CoM) onto the supporting surface [3]. The CoP will change in response to a perturbation of the body in space or of the supporting platform. Our lab employs a novel way to assess postural stability. Rather than making large, potentially fall- initiating pertur- bations, we make subtle translational perturbations that are at the edge of detectibility and that are buried within the range of a subject’s normal CoP sway path length (i.e., 2 mm rms, 20 mm range). We use iterative psychophysical testing procedures to determine the just detectible level (i.e., threshold) of acceleration at short 1, 4, or 16 mm anterior horizontal platform translations. To analyze differences in perturbations that were correctly detected versus those that were not, we needed an analysis tool take could work within the low signal (i.e. perturbation response) to noise (i.e., nor- mal sway) ratio inherent in our experiment. Thus, we wondered if it might be possible to take a sim- pler first-look approach to understanding postural stability — not by intensive back-calculations — but by treating the marker locations themselves as providing rich information about system state. Since the state of any system can be described by its eigenvalues, we felt that the set of marker positions might decompose into a set of physiologically relevant eigen-states. We thus proposed two hypotheses: H1: The most appropriate state variable for postural control is body configuration, approximated by measur- ing the position of each rigid element. The generic ODE model is given by: (1) Center of pressure (COP) describes variables of state only under a static condition. Dynamically, one should view COP as a mathematical projection of the body’s control signal. Mathematically, we could then assume that: ( 2) to indicate the functional relationship between the control (u) and its observable (COP). H2: A subject's ability to detect short, near threshold platform motion is affected by the state of the configura- tion variables at the time of movement. In particular, a subject is less likely to perceive motion if the body is moving toward its natural equilibrium. Part of a subject’s “sensation of movement” is based on detecting the neuro- muscular control response that is automatically initiated to maintain balance. If the body is moving toward equilibrium when the perturbation is inserted, less muscular action is required to maintain balance. Eigenposes: Using Principal Components to Describe Body Configuration for Analysis of Postural Control Dynamics Joseph D. Skufca, Erik M. Bollt, Rakesh Pilkar, Charles J. Robinson, Fellow IEEE I