無線加速度センサを用いた人の日常行動識別におけるデータ収集条件の影響評価(「BMI/BCI時代の心理学とVR」特集)

概要

論文の詳細を見る
In order to improve the daily activity classification with wireless accelerometers worn by people, we have evaluated how the daily activity classification performance depends on 1) the number of sensors and their bodily positions and 2) the sampling frequency. We have obtained the result that the highest classification performance of 10 kinds of daily activity is 89.9% when the accelerometers are installed in four positions (the both hands neck and both ankles) and SVM (Support Vector Machine) is used as a classifier. We have also obtained the result that the classification performance is 88.6% for the same accelerometer data except that they have been re-sampled from 50 Hz to 6.25 Hz. These results can be reflected to the wearable sensor design with less user load and longer working time.
日本バーチャルリアリティ学会の論文
2007-03-31