Head Gesture-Based on IMU Sensors: a Performance Comparison Between the Unimodal and Multimodal Approach

In this paper, we further develop systems used in head gesture recognition based on inertial sensors, previously developed and presented in other research papers. In the current research, we investigate the impact of increasing the number of inertial sensors as a condition to read head orientation with high accuracy. A set of eight predefined commands were evaluated with two systems that contain a single inertial sensor, respectively, three inertial sensors. The computational steps were done based on the Classical Machine Learning algorithms and Deep Learning algorithms. Each algorithm used in the study was trained based on raw data form of inertial signals. This study concluded that the increasing number of inertial sensors in the gestural recognition systems could improve the classification performance in the range of 10%-22%. Also, the average classification value calculated on ten predictive models is equal to 6.65%.