Action Quality Assessment Model Using Specialists' Gaze Location and Kinematics Data-Focusing on Evaluating Figure...

Area of Science:

• Computer Vision
• Sports Science
• Human-Computer Interaction

Background:

• Action Quality Assessment (AQA) in computer vision evaluates actions in videos, with applications in sports performance analysis.
• Previous AQA studies have not focused on predicting individual figure skating jump scores, a critical aspect for competitors.
• Figure skating videos contain extraneous information, yet human experts effectively filter relevant details.

Purpose of the Study:

• To investigate the eye movements of figure skating judges and skaters during jump evaluation.
• To develop a jump performance prediction model that incorporates specialist gaze data to reduce information.
• To enhance prediction accuracy by integrating kinematic features with visual attention data.

Main Methods:

• Recorded and analyzed eye movements of figure skating judges and skaters during jump evaluations.
• Developed a prediction model using video data, kinematic features, and specialist gaze locations.
• Compared the model's performance against human predictions and a baseline model.

Main Results:

• Identified differences in gaze patterns: skaters focused on the face, while judges focused on lower extremities.
• The prediction model achieved highest accuracy when incorporating gaze data from both judges and skaters.
• The proposed model significantly outperformed human predictions and the baseline model (RMSE: 0.775).

Conclusions:

• Specialist gaze location is a valuable feature for improving action quality assessment in figure skating.
• Integrating human expert knowledge (gaze data) with machine learning models enhances prediction accuracy.
• This approach offers a promising direction for objective and accurate sports performance evaluation.