Human Activity Recognition Based on Two-Dimensional Acoustic Arrays

Human activity recognition has been used in practical applications, gaming, monitoring of smart home, fire searching and rescuing, hospital patient management, etc. One or few acoustic sensors are used in conventional detection, which is required more features extracted from acoustic data for high accuracy. In this study, two-dimensional acoustic arrays are proposed for human activity recognition using convolutional neural networks. The experiments are employed to collect four activities (standing, sitting, falling, and walking) of volunteers. The features, as input of convolutional neural networks, are extracted from time- and frequency-domain data. The accuracy of the four activities is 97.5% for time-domain data. The results demonstrate the proposed method is effective in human activity recognition with high accuracy.