中文摘要：

基于智能手机的人体行为识别能用于健康监控和个人运动管理，针对不同用户携带手机的位置和习惯，分析基于手机传感器获取的三轴加速度信息，从人体不同位置思维行为数据中提取多种特征，优选出与行为相关度高且与手机位置相关度低的特征，构建三种决策树分类模型：（行为位置）矢量模型、位置一行为模型和行为模型，其中行为模型准确率最高；针对手机放置在三种不同位置的混合样本，其行为判断准确率为80．29％，耗时最短，能有效进行用户行为识别。

英文摘要：

Activity recognition by smartphone can be used for healthcare and sports management. People carry smartphones in many positions, such as the pocket of the trousers, hands or bags. This paper used accelerometer embedded in the smartphone to classify five activities, such as staying still, walking, running, going upstairs and downstairs. It used three machine learning algorithms for activity classification, and decision tree was the best way. Three models were constructed by decision tree： the （activity, position） vector model, the position-activity model and the activity model. Compared all these models, the activity model gain the highest accuracy and the least time-consuming, which can effectively identify human behavior.