Experimental analysis of accelerometer data for human activity recognition

Human activity recognition has received greater attention since last decade as datasets started showing up in literature. Some of these datasets have been developed in controlled laboratory environments or with practical considerations but with limited activities. There is still a need to generalize this effort to accommodate more human activities in practical environment either inside or outside. In this research, different components of human activity like feature extraction and classification of accelerometer data are addressed. We address feature extraction through filtering process and then classify these extracted features through machine learning tools. We show that tool like support vector machine is good enough for classification, but there is a need to improve feature extraction process using modern techniques like deep learning to automate feature extraction process. Experimental results are presented to compare experiments with expected optimum results.