| Summary: | 碩士 === 國立陽明大學 === 醫學工程研究所 === 101 === Sufficient physical activity can reduce the incidence of chronic diseases, obesity and depression. Nowadays, however, many people live without sufficient physical activity, and do not aware whether their daily activity is enough or not. If there is a system to record outdoor activities and remind the user whether their activity is sufficient or not, it will be helpful and easy to maintain sufficient physical activity in daily life. Activity level was used to quantify physical activity, and assist user to planning sport project. Therefore, our purpose is to use wearable sensor, as a sensing and reminding device cooperating with wireless system to help user maintain their physical activity. Before recording and reminding service, real-time activity recognition was the first step. In this work, we present a real-time activity recognition system by using wearable sensor. The accelerometer mounted in the front waist and back waist sensing, transmitting data to computer. The time series of raw data will be preprocess through the aggregation technique of jumping window. The raw data will be separated as gravity signal and body acceleration. Through feature extraction and threshold classification, real-time activity recognition is achieved. Finally, activity recognition can achieve 90% accuracy and activity level can achieve 80% accuracy. As a result, the system can apply for activity recognition and activity level.
|
|---|
