Simultaneous Recognition Algorithm of Human Activity and Phone Position Based on Multi-sensor Data Fusion

Abstract

For human activity recognition based on phone sensors, the position of the phone is an important factor of the recognition accuracy. To improve the recognition accuracy of behavioral activities and the position of the phone placed, this paper proposes a classification recognition algorithm based on accelerometer and gyroscope sensors. First, sensor data collected from seven different body positions are used as inputs to a deep stacked bidirectional long and short-term memory neural network; then the activity type and the phone position are used as labels to train the neural network for simultaneous recognition of human activity and phone position; finally, the performance of the proposed method is evaluated by cross-validation. The experimental results show that placing the phone on the waist and thigh achieves the highest recognition accuracy rate. The accuracy of the simultaneous recognition of activity and position is over 90%, which is 18% higher than existing algorithms.