Affiliations: [a] Department of Computer Science and Engineering, Wright State University, Dayton, OH 45435, USA. E-mail: tanvi.banerjee@wright.edu | [b] School of Nursing, University of California, Los Angeles, CA, USA. E-mail: m.yefimova@ucla.edu | [c] School of Electrical Engineering, University of Missouri, Columbia, MO 65211, USA. E-mails: kellerj@missouri.edu, skubicm@missouri.edu | [d] School of Nursing, Azusa Pacific University, Azusa, CA, USA. E-mail: dwoods@apu.edu | [e] School of Nursing, University of Missouri, Columbia, Columbia, MO, USA. E-mail: rantzm@health.missouri.edu
Abstract: We describe case studies of clinically significant changes in sedentary behavior of older adults captured with a novel computer vision algorithm for depth data. An unobtrusive Microsoft Kinect sensor continuously recorded older adults’ activity in the primary living spaces of TigerPlace apartments. Using the depth data from a period of ten months, we develop a context aware algorithm to detect person-specific postural changes (sit-to-stand and stand-to-sit events) that define sedentary behavior. The robustness of our algorithm was validated over 33,120 minutes of data for 5 residents against manual analysis of raw depth data as the ground truth, with a strong correlation (r=0.937, p<0.001) and mean error of 17 minutes/day. Our findings are highlighted in two case studies of sedentary activity and its relationship to clinical assessments of functional decline. Our findings show strong potential for future research towards a generalizable platform to automatically study sedentary behavior patterns with an in-home activity monitoring system.
