Abstract—Postural sway is a well known measure of postural stability in the elderly. Sway measurements, however, are typically made using expensive equipment in a laboratory. We report on efforts to make clinically significant and quantitative measurements of postural sway in a community center with a single un-calibrated video camera. Results indicate that simple tracking technologies can capture some aspects of sway in a community center in a way that is perceptually accurate and capable of distinguishing expert-assigned levels of balance performance in an elderly, balance impaired cohort. I. INTRODUCTION ncreased postural sway as recorded during quiet standing has been found to correlate both with ageing [1,2] and an increased likelihood of falling [3,4,5]. Traditionally, postural sway has been measured by recording an individual’s center of pressure (COP) with a posturographic platform [1,6,7] or by inferring center of mass trajectories using kinematic estimates provided by commercial motion tracking devices [8]. Both measurement systems are extremely accurate, but they are also expensive, not readily portable, and require the subject to wear potentially cumbersome markers. As new technologies develop, however, it is becoming increasingly possible to measure movements like postural sway more cheaply, in a wider range of situations and environments, and without the need for on-person markers. Accelerometers, for example, have proven to accurately detect abnormal postural sway in the elderly [8]. Unfortunately, accelerometers still require individuals to wear external sensors. Movement data recorded by video cameras provide a potential alternative for affordable capture of human motion in a wide range of settings, possibly without markers. Markerless motion capture, in fact, has received a great deal of recent attention from the computer vision and biomechanics community [10,11], and, in some circumstances, has proven to yield kinematic reconstructions comparable to those from high cost tools, even when based on single camera views [12]. The clinical applications of such tracking technologies, however, have not been extensively explored until recently. Markerless video tools Manuscript received April 7, 2008. This work was supported in part by the Vancouver Foundation, the NSF RERC on Universal Design and the Intramural Program of the NIH/NIMH. 1 Sonya Allin and Alex Mihailidis are with the Intelligent Assistive Technologies and Systems Laboratory, University of Toronto, Toronto, ON (email: s.allin at utoronto.ca, alex.mihailidis at utoronto.ca). 2 Cheryl Beach is with the Vancouver Island Health Authority, BC. 3 Andrew Mitz is with the National Institutes of Health, Bethesda, MD. have be used for posture identification for ergonomic workplace analysis [13], biomechanical analysis of sit to stand movements [14], and as input to assistive functional electric stimulation (FES) devices [15]. The purpose of this paper is to determine if a single un- calibrated camera in a community center can measure clinically meaningful statistics of postural sway among an elderly, balance impaired cohort. We are not using dense or multi-camera stereo methods as in [11]; our focus, rather, is on the clinical utility of coarse camera-based measurements that can be made robustly and quickly. Our work is most closely related to [16], wherein a single camera was used to distinguish between standing conditions (eyes open, eyes closed, etc.) among a population of young subjects. Our work builds upon this research in several important ways. First, we explore the use of a camera positioned in front of subjects as opposed to laterally with respect to the subject. This provides access to medio-lateral sway measurements, which have been shown to increase with age [4]. Second, we explore the use of un-calibrated cameras in a real world scenario that involves balance impaired elders. Finally, we validate accuracy of measurements by determining if they capture expert determined levels of functional balance performance in the elders we record. The opinions of human experts are the common “gold standards” for clinical measurement of functional balance in community centers, so we use these opinions as our “gold standards” as well. II. METHODS A. Experimental Paradigm TABLE I SUBJECT DEMOGRAPHICS ID Age Total BBS Postural Sway BBS Medical History 1 81 37 (49) 3 (4) Arthritis, stroke, knee replacement 2 86 3 (7) 0 (0) Arthritis, macular degeneration 3 77 41 (44) 4 (4) Arthritis, Parkinson’s, hip replacement 4 85 23 (26) 0 (0) Stroke 5 87 19 (22) 1 (3) Arthritis, diabetes Table 1. The total BBS is the summed score achieved across all “items” on the Berg Balance Scale (range is 0 to 56). The Postural Sway BBS score is the score received on the Berg item which demands that subjects stand, feet together, for 1 minute (range is 0 to 4). Scores outside of parentheses were recorded BEFORE the balance classes. Scores in parentheses were recorded AFTER classes were completed. Video based analysis of standing balance in a community center Sonya J. Allin 1 , Cheryl Beach 2 , Andrew Mitz 3 , and Alex Mihailidis 1 I