Abstract: Activity recognition is a fundamental and challenging topic in context-aware applications. However, the current research on activity recognition cannot obtain satisfactory results by considering the complexity and diversity of human activity. To solve this problem, a novel method for activity recognition is proposed. The method extracts feature vector from three-dimensional acceleration sensor and gyro sensor of smart-phone, then construct recognition models based on four typical statistical methods（namely, K-nearest neighbor algorithm, support vector machines, naive bayes networks and AdaBoost） respectively, and finally select the optimal recognition model according to model evaluation and comparison. Experimental results show that the proposed method achieved good performance and the accuracy rate of activity recognition is 92%.

Key words: activity recognition, acceleration, gyroscope, statistical learning model

摘要： 人体活动识别是上下文感知系统及其应用中一个具有挑战性的研究问题。目前，关于人体活动识别的研究主要使用一些基于监督学习或半监督学习的统计方法来构建识别模型。然而，考虑到识别活动类型本身具有的复杂性和多样性，当前的人体活动识别系统不能取得较好的识别效果。针对这一问题，通过智能手机的三维加速度和陀螺仪传感器信息来提取人体活动的特征向量，选择四种典型的统计学习方法（分别是K-近邻算法、支持向量机、朴素贝叶斯网络以及基于朴素贝叶斯网络的AdaBoost算法）分别创建人体活动的识别模型，最后通过模型决策得到最优的人体活动识别模型。实验结果表明，通过模型决策选择的识别模型对人体活动识别准确率达到92%，取得很好的识别效果。

关键词: 活动识别, 加速度, 陀螺仪, 统计学习模型