CNN based Identifying Human Activity using Smartphone Sensors

Human Activity Recognition has gained greater emphasize in the last few years due to its widespread applicability and psychological curiosity. This system can be adopted in innumerable applications, like healthcare monitoring systems, surveillance systems, and so on. Smart-phones have built-in multifunctional sensors like as accelerometers and gyroscopes that provide useful sensory data when participants perform daily activities thus helping in HAR activity. Highly efficient features are extracted from this sensor data and techniques like denoising, normalization and segmentation are used to reduce noise and extract valuable feature vectors. Prior research showed that deep learning methods like recurrent neural networks and one-dimensional convolution networks provide excellent results in activity recognition tasks. In this paper, an ensemble model of CNN and SVM is proposed to further improve the accuracy and provide a robust model. Experimental methods are tested on UCI-HAR dataset and compared with other state-of-the-art methods like LSTM, CNN-LSTM, and Conv LSTM.