Human Activity Recognition from Environmental Background Sounds for Wireless Sensor Networks

Sound feature extraction Mel frequency cepstral coefficients (MFCC) and classification dynamic time warping (DTW) algorithms are applied to recognizing the background sounds in the human daily activities. Applying these algorithms to fourteen typical daily activity sounds, average recognition accuracy of 92.5% can be achieved. In these algorithms, how two parameters (i.e., Mel filters number and frame-to-frame overlap) affect system's calculation burden and accuracy is also investigated. By adjusting these two parameters to a suitable combination, the calculation burden can be reduced by 61.6% while maintaining the system's average accuracy rate at approximate 90%. This is promising for future integrating with other sensor(s) to fulfill daily activity recognition work by using power aware wireless sensor networks (WSN) system.