Use the Direction of the Combined Vector of Accelerometers to Classify Daily Activities

Abstract

Human Activity Recognition (HAR) is important for human physical fitness. Wearable sensor-based HAR uses accelerometers (A), gyroscopes (G), and other sensors to collect human motion data. To obtain a longer battery life for the wearable device and conduct long-term monitoring of human daily activities, it is necessary to adopt a method that can extract features from signals of sensors efficiently. In this paper, we study the extraction method of accelerometer data and combine the three axes of accelerometers into a direction ax. Then encode the extracted directions into numbers 1–62. In different activities, the distribution of directions is significantly different. Using Artificial Neural Network (ANN), K-nearest neighbor (KNN), Random Forest (RF), and Support Vector Machine (SVM) as classifiers, and the direction distributions as features input to classifiers, the classification accuracy up to 99.3%, in the best case.