Bayesian Intention Inference Based on Human Visual Signal

In the robot-assisted technology designed for the disabled, it is very challenging to infer the potential intentions based on accurate eye movements. This paper proposes a framework to capture the user's eye movement characteristics, infer their implicit intentions, and realize the interaction between humans and assistive robots. The framework uses the I-DT algorithm to extract the intentional gaze points and uses the DBSCAN algorithm to cluster their intentional gaze points to identify the subject's area of interest and the corresponding objects under the area of interest, thereby building a small intent knowledge base. Using this as a priori information, when the intention is unknown, the Naive Bayes model is used to infer the intention of the eye movement event. The empirical results of 9 subjects using the Pupil Core eye tracker show that the frame accuracy rate can reach 86.7%, indicating that the proposed eye movement event recognition and intention inference have certain validity and accuracy.