Key points are not available for this paper at this time.
The smartphone has become very popular and it is very useful in our daily life. A smartphone is equipped with a variety of sensors, which makes it very suitable in replacing other types of wearable sensors for activity recognition of the user. Many researches have been carried out with this subject. However, most of them have to confine the smartphone to a certain position or even orientation. For example, the smartphone might need to be placed in the front pocket of pants to make the right recognition. In this research, we hope to relieve this limitation and to develop a smartphone APP for activity recognition. Activity recognition can be done in two phases. In the first phase, gyroscope data are used to decide the position of the smartphone by the Support Vector Machine (SVM). It could be of three categories: (1) front pockets of pants, (2) back pockets of pants, and (3) shirt pockets or bags. Different classifiers are used for smartphone with different position categories. SVMs are trained for different position categories using accelerometer data. Each of them can recognize activities of walk, run, up-stairs and down-stairs. The developed APP can recognize activities of the smartphone user in real time and accumulate the times for different activities. The user can check the results through the APP at any time.
Hsu et al. (Tue,) studied this question.