IMU have been widely adopted for body-motion tracking in industrial applications and product designs, thanks to its portability, low latency, and cost-effectiveness in 3D pose estimation.
In this thesis, a dual-IMU-based system was developed to infer fingertip position from the inertial data of two IMU sensors when the user approaches specific directions or positions. The proposed solution will explore different types of deep-learning networks to evaluate the tracking performance.
The inferred fingertip 3D trajectories could not only potentially serve as a feedback control signal for the future prosthetic robotic arm, but also pave the way for seamless integration of finger-level position estimation into wearable and assistive devices.