Time Series Feature Extraction For Head Gesture Recognition: Considerations Toward HCI Applications
Ionuţ-Cristian Severin · 2020
In this paper, it was further developing the head gesture recognition system based on the inertial sensor as a necessary prerequisite in any HCI or healthcare application. While in one previous research, it was got a good result equal to 80% head classification accuracy, in this work, the main aim was to increase the classification rate using a different approach than the previous one. For this paper, a database with records performed from 7 persons was used and 8 head activities were classified. The increase of the performances for this research was done by implementing two specific methods applied to the time series. The first method was based on a signal processing approach - the extraction of the trend and the filtering, with the help of the median filter, composed this method. The second approach was based on the extraction of 16 new features from the proposed head gestures recognition time database. In the end, it was constructed 9 predictive models for the head gestures classification and evaluation. After applying the proposed methods, the evaluation part was done using the K-Fold method. This thing shows that the first proposed approach was got an accuracy equal to 91% while the second one was got a classification rate equal to 100%. Based on the proposed methods, in this paper, the classification performance grew up with 10% and respectively, 20% compared with the previous research.