Human action recognition based on a single acceleration sensor
Yi Wu · Journal of Physics Conference Series · 2019
Abstract In this paper, based on the acceleration signal acquired by a single triaxial acceleration sensor worn at the bottom of the finger, an algorithm for recognizing the daily movement behavior of the human body is proposed. The algorithm converts the three-axis acceleration into the resultant acceleration, horizontal acceleration and vertical acceleration, and extracts the features from them. The 37 features are sorted by the F-score as a measure of the importance of the feature, and a subset of features that meet the application requirements is found. Model training is performed on the feature subset using multi-class SVM to obtain the final classification model. Experiments show that the model has high recognition accuracy and the accuracy can reach 93.17%.