Overview of Research

To cut health care costs, it is increasingly common to discharge patients from rehabilitation facilities after short stays. This trend is observable after stroke, traumatic brain injury, and orthopedic surgery. Although there is evidence to suggest that rehabilitation at home can yield functional outcomes equivalent to those achieved in the hospital, there is also evidence to suggest that rehabilitation at home may increase risk of falls, for example. This project seeks to develop affordable automated tools to unobtrusively perform longitudinal, quantitative analyses of functional balance so as to better understand rehabilitation in environments like homes and workplaces, and to guarantee it is safe and effective.

The research objectives for this project are to:

develop affordable and home appropriate computer vision tools that can track the locations of elder's bodies as they rise from chairs, climb stairs or walk;
translate kinematics observed during movements onto assessments of motor quality that are consistent with leading clinical assessments of balance;
determine if affordable measurements can produce stability analyses comparable to those made with more sophisticated clinical devices, like commercial motion capture systems or force plates.

There are many ways to track motion with video data. We focus primarily on kinematic trackers; these represent the human body as assemblages of multiple three dimensional body parts, like a head, arms and legs. Tracking is done with stereo calibrated pairs of USB cameras, which makes it possible to combine recorded tracks across cameras to reconstruct motion in three dimensions. In Figure 1 is an example of a balance impaired individual whose kinematics are being tracked she rises from a chair in a community clinic. Tracking parameterizes the motion of her head, torso and legs. Several mobility statistics can be produced with the resulting tracks, including the angle of her torso over the duration of the sit-to-stand; this is illustrated in Figure 1. Many of these statistics, moreover, have been shown to correlate with balance assessments that are determined by a human expert. In Figure 2, for example, velocity of individuals’ heads is illustrated, as are their expert determined functional balance scores. It is evident that the less smooth this velocity, the lower the functional score. We hope to identify more features like this as our research continues, so that we may make automated balance assessments of behaviours, like rising from chairs, in homes or offices.

Figure 1. A balance impaired individual being tracked in three dimensions as she rises from a chair. At the base of the figure is the angle of the individual’s torso, reconstructed using USB camera data, overtime. This angle has been smoothed and the time points corresponding to images indicated with the dots. [click to enlarge image]

Figure 2. Head velocity of five individuals during a sit-to-stand transition. Expert determined functional scores on the Berg Balance Scale for each individual are given in red. Peaks in the velocity profiles are marked with read dots. It is evident that the less smooth the velocity profile (i.e., the more peaks in the profiles), the lower the functional score. [click to enlarge image]

Figure 3. In preliminary studies, reconstructed motion perceived with web cameras correlates strongly with motion perceived by a commercial motion capture system. Computed torso flexion for an elderly individual as she rises from a chair is at the base of the figure. The motion capture measure of the flexion is in blue. The raw measure of flexion from the USB cameras is in green, and a smoothed version is in red. [click to enlarge image]

Funding Sources

Natural Sciences and Engineering Research Council of Canada (NSERC) Post-Doctoral Fellowship

Research Team

Elham Dolatabadi, University of Toronto

Sonya Allin, Toronto Rehabilitation Institute

Alex Mihailidis, University of Toronto

Brian Maki, Sunnybrook Health Sciences Centre

Karl Zabjek, University of Toronto