Silva Joana, Monteiro Miguel, Sousa Filipe
Fraunhofer Portugal AICOS, Porto, Portugal.
Faculdade de Engenharia da Universidade do Porto, Porto, Portugal.
Stud Health Technol Inform. 2014;200:101-4.
Monitoring human physical activity has become an important research area and is essential to evaluate the degree of functional performance and general level of activity of a person. The discrimination of daily living activities can be implemented with machine learning techniques. A public dataset provided during the European Symposium on Artificial Neural Networks 2013, with time and frequency domain features extracted from raw signals of the smartphone inertial sensors, was used to implement and evaluate an activity classifier. Using a decision tree classifier, an accuracy of 86% was achieved for the classification of walk, climb stairs, stand, sit, and lay down. The results obtained suggest that the smartphone's inertial sensors could be used for an accurate physical activity classification even with real-time requirements.
监测人类身体活动已成为一个重要的研究领域，对于评估一个人的功能表现程度和总体活动水平至关重要。日常生活活动的识别可以通过机器学习技术来实现。利用在2013年欧洲人工神经网络研讨会上提供的一个公共数据集，该数据集具有从智能手机惯性传感器的原始信号中提取的时域和频域特征，用于实现和评估一个活动分类器。使用决策树分类器，在对步行、爬楼梯、站立、坐下和躺下的分类中达到了86%的准确率。所获得的结果表明，即使有实时要求，智能手机的惯性传感器也可用于准确的身体活动分类。
