Vision-Based Attentiveness Determination Using Scalable HMM Based on Relevance Theory.

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Robot Interaction

Background:

• Effective human-robot interaction relies on a robot's ability to gauge and respond to human attention.
• Existing robot attention models struggle with noisy data and adapting to dynamic participant numbers, leading to attention instability.
• Previous approaches lack adaptability, failing to adjust robot attention when the number of people changes.

Purpose of the Study:

• To introduce a novel algorithm for determining and prioritizing human attentiveness in human-robot interaction.
• To address the limitations of existing models, specifically their susceptibility to noise and inability to adapt to varying numbers of participants.
• To enhance the robot's ability to focus its attention effectively in dynamic social environments.

Main Methods:

• Development of the Scalable Hidden Markov Model (Scalable HMM), an algorithm based on relevance theory.
• Scalable HMM features a scalable number of states and observations, enabling adaptation to a changing number of participants.
• Online adaptability for state transition probabilities allows real-time adjustments to participant number variations.

Main Results:

• The Scalable HMM algorithm was tested on 7567 frames of human actions in various scenarios.
• Achieved a 76% detection rate for identifying the most attentive person.
• Demonstrated over 75% accuracy in prioritizing attention among a varying number of participants, showing a 20% improvement over existing methods.

Conclusions:

• The Scalable HMM provides a robust and adaptable solution for robot attention determination in human-robot interaction.
• This model effectively handles noisy observations and dynamic changes in the number of participants.
• Scalable HMM significantly improves the accuracy and stability of robot attention prioritization compared to prior approaches.