Application of dynamic time warping based algorithm for motion similarity evaluation and development of a home-based interactive fall intervention system

Abstract

The global healthcare system is under enormous pressure because of rapid trend in population aging and falls cause substantial health and safety problem among elderly people. Although the effectiveness of conventionally prescribed exercises on fall risk reduction and motor function improvement has been well accepted, many elderly people still face difficulty engaging in such exercises for a long period of time due to passive nature of the conventional exercise, inconvenience, cost and lack of skill. One solution to address these barriers is to develop an interactive, home-based, low-cost gamified system for fall intervention by the mode of virtual coaching. Because there is no real coach in such a system and older people are not familiar with most of the new technologies, scientific algorithm for motion similarity evaluation, and good technology acceptance of such a system are very critical to make it be a practical application at home environment.

The major objective of this study is to provide effective home-based virtual coaching offering similar outcome to those obtained from conventional fall intervention programs under the supervision of a real coach. The algorithm for performance evaluation should not only take account of different speeds (movement paces) between an elderly user and a virtual coach, but also provide meaningful overall performance evaluation and informative feedback to achieve just-in-time coaching. In this study, we proposed a dynamic time warping (DTW) based algorithm for evaluation of motion similarity between motion time-series data from an individual user and a virtual coach. The effectiveness of the proposed algorithm was validated in a follow-up experiment, in which the performances of 21 subjects playing a Tai-chi exergame were evaluated by our algorithm and domain experts respectively. The result showed that the evaluation from our algorithm has a strong linear relationship ($R^2$ = 0.80) with experts’ evaluation, which suggested that our algorithm could be effectively used for performance evaluation of an individual when performing home-based rehabilitation exercises.

We also did an exploratory study on applying the validated algorithm into developing a low-cost, home-based prototype system for interactive fall intervention utilizing exergame technology and a markerless Kinect. A further experiment on evaluating the technology acceptance of this prototype system was carried out among 41 intended elderly users. The result showed that elderly people expressed positive attitude (Attitude = 4.29 out of 5) to use our system in the future (Behavior intention = 4.08 out of 5). However, based on their evaluation (Effort expectancy = 3.07 out of 5), we also found that the system needed to be further improved especially in terms of ease of use (e.g., some motions are difficult to follow

some instructions are difficult to understand).

The proposed algorithm for motion similarity evaluation was implemented and validated in the real exergame application based on Tai Chi, which can be easily extended to evaluate user’s performance on other exercises or sports. The findings from acceptance evaluation will guide us to further improve our system for practical applications on home healthcare in the future. The final released system could be promising to minimize the risk of falling, improve the life quality of older people and reduce the cost of public healthcare system.