Cross-People Motion Activity Recognition
Zhongtang Zhao, Yiqiang Chen, Junfa Liu and Mingjie Liu
In smart phone based motion activity recognition which is the kernel technology of healthcare, the model learned from a specific person often doesn’t satisfy a new user whose activity samples are not included in the training set. The traditional method to solve this problem is to build a specific model for the new user. But collecting the new user’s labeled samples needs huge manual efforts. In this paper, our algorithm learns a binary decision tree model for one person from his labeled samples, transfers its structure to another person and automatically adapts its non-determinate nodes with the unlabeled samples of the new person, thus accomplishes the cross-people knowledge transfer task. Our algorithm is tested on the real-world data set and the results show that it outperforms the traditional methods.