Hand 3D Trajectory Estimation for BCI Application

The state of art Brain-computer interface (BCI) utilized discrete or model-based control strategies for external device control. However, for efficient and seamless control a continuous control strategy is required. In order to achieve this continuous estimation of control parameters is required with minimum delay. It will improve the performance as well as acceptability of the mind-controlled prosthesis, exoskeleton and robotic arm among the uses. In this research paper, an attempt had been made to estimate the human hand trajectory in 3D space using multichannel electroencephalogram (EEG) signals. The proposed model utilized a time-delayed multi-input multi-out neural network to estimate the trajectories in a continuous manner. The developed model is well suited for control applications as it generates a high-density of estimated trajectory stream. The developed model has been tested over the dataset of 12 subjects for different frequency ranges/bands of EEG signal. The developed model shows the best estimation accuracy as 0.638±0.030 and consistency of estimation as 0.654±0.030, if the entire frequency range of the EEG signal has been utilized. The developed model depicted better performance if utilized for trajectory estimation in 2D space rather than 3D space. The developed model can be directly utilized for planer robot control or any upper limb assistive and rehabilitative device with 2DoF.