Mobile health monitoring is driving many smart health innovations. In this research, we focus on the daily physical activity and fall risk monitoring, for intelligent activity tracking and fall risk detection purpose. Physical activity is of a tremendous value to cardiac and brain health, and has been widely agreed as a metric to improve human health, foster healthy lifestyle, and mine the impacts on diverse health issues. The risk of death can actually be lowered by 20 to 30% with moderate activities for one and half an hour per week (WHO). Besides, fall risk detection is also of huge importance since each year, there are 29% of older adults having around 29 million falls and causing 7 million fall injuries (CDC). The challenge we target is how to accurately recognize diverse physical activities and fall patterns from the biomechanical dynamics sensed by the phone. More specifically, the phone sensor is leveraged to provide real-time always-on streaming of biomechanical dynamics. Then, the data is learned by a deep learning framework (long short-term memory) for activity type and fall pattern recognition. To consider real-world scenarios, the users can put the phone in either left or write waist pockets. The users have performed 9 different kinds of activities such as walking, standing up, going upstairs and jumping, as well as 8 different kinds of fall patterns such as falling backward, sitting and falling, and falling forward with protection. A total of 11,770 activities performed by 30 subjects aging from 18 to 60 have been used to evaluate the framework, yielding an accuracy as high as 95.4%. This encouraging result shows the potential of the mobile monitor in real-world application scenarios.
Read full abstract