无监督特征学习的人体活动识别

针对人的局限性可能会导致在提取特征中丢失重要信息,从而影响最终的识别效果问题,提出无监督特征学习技术的惯性传感器特征提取方法。其核心思想是使用无监督特征学习方法学习多个特征映射,再将所有特征映射拼接起来形成最终的特征计算方法。其优点是不会造成重要信息的损失,而且可以显著减少所使用的无监督特征学习模型的规模。为了验证所提出的特征提取方法在活动识别中的有效性,运用一个公开的活动识别数据集,使用三种常用无监督模型进行特征提取,并使用支持向量机进行活动识别。实验结果表明,特征提取方法取得了良好的效果,与其他方法相比具有一定的优势。

Unsupervised feature learning for human activity recognition

Human activity recognition using smartphone inertial sensors is currently an important research direction in the field of pervasive computing. Feature representation has a significant impact on human activity recognition. While the common used features, including time domain features and frequency domain features, heavily rely on the specific domain knowledge and are designed by hand, which may cause the loss of important information, thus affecting the classification results. To solve these problems, this paper proposed a feature extraction method based on unsupervised feature learning techniques. Firstly, we use unsupervised feature learning method to learn multiple feature maps, then concatenate them together. This method can avoid the loss of important information, and also can significantly reduce the scale of unsupervised feature learning model used. To evaluate the proposed method, we perform experiments on a public human activity recognition dataset, using three commonly used unsupervised feature learning models, and finally using support vector machines to classify activities. The results show that the proposed feature extraction method achieves good results, and has certain advantages compared with other methods.