A framework of imitative behavior analysis for animal exercise courses via human pose estimation.

In contemporary performance education, the Animal Exercise course is one of the core training modules for developing imitative behavior. Typically, instructors facilitate this process through guided demonstrations and task-based instruction, encouraging students to engage in both imitation and creative exploration. The pedagogical approach is therefore characterized by active student participation and a strong emphasis on experiential, practice-oriented learning. However, assessment in Animal Exercise courses still relies primarily on instructors' subjective judgment, resulting in inconsistent and non-standardized evaluations. This hinders students' ability to identify skill deficiencies and improve their course performance. To address this challenge, we propose a quantitative framework for evaluating imitative behavior using pose estimation, termed Human Pose Estimation-Imitative Behavior Analysis (HPE-IBA). Using this framework, we employ a standard RGB camera to collect motion data from both students and gorillas, extract three-dimensional joint coordinates, and compute dynamic joint angles with MediaPipe. We then apply correlation analysis to identify weakly correlated features and core joints, followed by two-way ANOVA to examine the effects of training status and gender on students' imitation performance. Analysis of chest-beating and walking imitation reveals a statistically significant interaction between training status and gender (p< 0.01), primarily reflected in joint patterns such as the right elbow and right knee. The proposed framework not only enhances the application of pose estimation in acting education but also provides a foundation for broader applications in performance-based motion analysis.