Development of the Estimation Model for Intentions to Move Based on Gaze and Face Information with 1DCNN-LSTM and Evaluation of Electric Wheelchair Driving
Sho Higa, Kôji Yamada, Shihoko Kamisato · 2021
Gaze interaction depicts a lightweight and fast method for hands-free control of machines, as it is not susceptible to paralysis, has been made to use it as an interface for electric wheelchairs. However, there is a problem called the Midas Touch Problem, in which unconscious eye movements are also treated as input operations. In this paper, we propose the estimating model for Intentions to move based on gaze and face information to develop an electric wheelchair that can be operated by the natural gaze and face movements of users. The proposed model is consisted of 1D-CNN and LSTM layers and uses time-series data of gaze and face movements as input. As a result of the experiment, the highest estimation accuracy was obtained when input the time-series data of 2 seconds to the proposed model.