Human activity recognition based on smart phone’s 3-axis acceleration sensor

Abstract

With the rapid development of smartphone, human activity recognition based on acceleration sensors attracts much attention in the academic and industry recently. However, the recognition accuracy is not ideal due to the diversity of human activities and other environmental factors. A real-time user activities monitoring system is developed on android, and comparison of several feature extraction and classification algorithms is carried out. Based on the monitoring system, a feature called (TF4+FFT10) is proposed. Experiment result shows that the recognition accuracy rate of feature (TF4+FFT10) with the adopted KNN algorithm is 98.6%.