Analysis of Multi-Sensor Fusion for Mobile and Wearable Sensor Based Human Activity Recognition

Abstract

Sensor-based human activity monitoring and detection have become an emerging field of intense research and development in recent years due to its immense applications in wide area of human endeavors. Human activity recognition integrates diverse sensors with machine learning algorithms to provide contextual information on relative activity details for health-related feedbacks and lifestyles changes. However, there are varieties of sensors for implementing human activity recognition with diverse capabilities and types of activities they provide best performances. Also, the diverse nature of human activities and nature in which they are performed by the individual makes them challenging to recognize. Therefore, determining the impact of these sensors in human activity recognition using machine learning techniques are of immense advantages in human activity monitoring and detection. The objective of this paper is to comprehensively evaluate the performance of single and multi-sensor fusion for human activity recognition using accelerometer and gyroscope sensors. Firstly, the performances of these sensors were extensively analyzed individually using seven classification algorithms. Secondly, we conducted a comprehensive experimental evaluation of sensor fusion attached on the same location and on different locations of the body. The extensive evaluation with 10-fold cross validation demonstrates that highest average F-measures for single sensor and fusion are 0.908 and 0.938 with Random Forest and Voting ensemble algorithms respectively. Furthermore, the fusion of heterogeneous sensors attached to different locations of the body shows Chest and Hip sensors fusion achieves an average F-measure of 0.942 and classification accuracy of 94.23% using Random Forest algorithm. The outcome of our experimental evaluation shows the significant impact of multi-sensor fusion for human activity monitoring and detection.