Sensor Fusion for Effective Hand Motion Detection
Abstract
Human hands have a critical role in many human activities. There are many applications in the field of human computer interaction where it is necessary to monitor human hand motions. Different technologies have been employed to track the human hand motion. However, many of those approaches did not work well in real life situations. This research focused on tracking the human hand motion so that it can be used regardless of location or environmental lightening. MEMS (Micro-Electro-Mechanical sensors) inertial and magnetic sensors were used in this study for tracking the palm and the thumb motions. In this dissertation a sensor fusion algorithm was proposed to track the three-dimensional rotational motion of the human hand. The proposed sensor fusion algorithm was a quaternion-based Kalman filter. Two optimization methods (Gradient Descent and Newton-Gauss) were utilized to estimate the corresponding quaternion vector using the observation vector in the Kalman filtering process. Two sensor units were attached on a glove on the palm and thumb sections to record the motion related data. As the articulation of the remaining four fingers (index, middle, ring and pinky) to the palm are more restricted than the thumb, the applied approach for the thumb can be applicable for other fingers. It is challenging to compute the rotational motion using MESMS sensors with acceptable accuracy because the MEMS sensors are prone to different types of systematic and stochastic errors. To calculate reliable results from these sensors, it was necessary to compensate for all types of error. The calibration process was performed to compensate for systematic errors. The sensors stochastic errors were compensated in the sensor fusion process. To evaluate the system performance, three experiments were carried out. Statistical analysis was performed on the experimental data. Statistical analysis showed that orientation provided by the sensor fusion algorithm is accurate and reliable. It was also observed that both optimization techniques performed similarly, although the Newton-Gauss approach converged faster. Additional results are found in this dissertation.
Subject Area
Electrical engineering
Recommended Citation
Abyarjoo, Fatemeh, "Sensor Fusion for Effective Hand Motion Detection" (2015). ProQuest ETD Collection for FIU. AAI10002853.
https://digitalcommons.fiu.edu/dissertations/AAI10002853