Recognizing Human Activities in Real-Time Using Mobile Phone Sensors

To overcome the defects that previous research cannot recognize human activities accurately in real-time, we proposed a novel method, which collects data from the accelerator and gyroscope on a mobile phone, and then extracts features of both time domain and frequency domain. These features are used to learn random forest models offline, which make our mobile app can recognize human activities accurately online in real-time. Verified by theoretical analysis and a large number of contrast experiments, the recognition is rapid and accurate on mobile phones with accuracy at 97 %.