Activity Recognition Based on Deep Learning and Android Software

This project transplants the deep learning framework to the android platform. It combines Deep learning based on Tensor Flow (TF) with Android software to recognize people's activities. Provided by android sensors, dataset is in the form of float arrays. And then we used them to train TF model with the data that was tagged by the different actions. Then we transplant the model into android to achieve activity recognition in mobile phone. In this project, six activities are recognized, including downstairs, upstairs, jogging, walking, sitting and standing. Finally, the probability of each activity will be seen in an android mobile phone. According to the results of experiments, almost all of six types of activities can be recognized successfully.