Intelligent Eye-Controlled Electric Wheelchair Based on Estimating Visual Intentions Using One-Dimensional...

Area of Science:

• Biomedical Engineering
• Computer Science
• Human-Computer Interaction

Background:

• Gaze-based control of electric wheelchairs faces the "Midas touch problem", where unintended eye movements are misinterpreted as commands.
• Accurate classification of visual intentions is crucial for effective and safe human-computer interaction in assistive technologies.

Purpose of the Study:

• To develop a real-time deep learning model for estimating user visual intention in electric wheelchair operation.
• To integrate this intention estimation model with the gaze dwell time method for an improved wheelchair control system.

Main Methods:

• A 1DCNN-LSTM deep learning model was developed to estimate visual intention using feature vectors from eye movement, head movement, and fixation point distance.
• The model was evaluated by classifying four types of visual intentions.
• Driving experiments were conducted with an electric wheelchair incorporating the proposed model.

Main Results:

• The proposed 1DCNN-LSTM model achieved the highest accuracy in classifying visual intentions compared to other evaluated models.
• Electric wheelchair driving experiments demonstrated reduced user effort and improved operability with the new system.
• The study confirmed that learning time-series patterns from eye and head movement data enhances visual intention estimation.

Conclusions:

• The developed deep learning model accurately estimates user visual intentions for electric wheelchair control.
• The integration of intention estimation and gaze dwell time significantly improves wheelchair operability and user experience.
• This approach offers a promising solution to the challenges posed by the "Midas touch problem" in gaze-controlled systems.