中文摘要：

通过手机内置加速度传感器数据对人类日常行为进行识别具有便捷、实时、无干扰的优点，为了提高识别的准确率和稳定性，提出一种基于加速度特征稀疏矩阵字典的分类方法识别行为。从不同行为的多个训练样本构造出一个过完备字典，基于该字典通过求解最小z，范数得到待识别样本的稀疏系数，根据稀疏系数计算待识别样本对应不同行为的残差并选取最小值对应的行为作为分类结果。实验表明该方法识别手机用户日常行为可以达到84．93％的准确率，高于传统的决策树和BP神经网络算法的分类准确率，且分类稳定性也优于传统分类方法。

英文摘要：

Recognizing human daily behaviors with the data sampled from mobile phones＇ buih-in accelerometer is conve- nient,real-time, and non-interferential. In order to improve the accuracy and stability of recognition, this paper proposed a classification method based on sparse matrix dictionary of acceleration features. It constructed an over-comprehensive dictionary by training samples of different kinds of activities firstly. And then it calculated the sparse coefficient for samples to be tested by solving the minimum 11 norm. At last,it calculated residual values according to the activities and selected the minimum one as the indicator to obtain the classification results. Experiments show. that this method can reach a recognition rate of 84.93% for recognizing mobile phone users＇ daily behaviors ,which is higher than the recognition rate obtained by traditional classifica- tion algorithms, such as decision tree and BP neural network. At the same time, the stability of classification of this method is also superior to the traditional classifications.