Real-Time Person Recognition Using MoveNet for Pose-Based Identification

Person recognition is a critical task in computer vision. This study suggests person recognition based on skeletal keypoints extracted from pose estimation models. Using MoveNet, it is explored how effectively a system can identify people by comparing their normalized body proportions. Experiments with a dataset of ten participants reveal that recognition accuracy is highest when the candidate pool is small and individuals possess distinct body types reaching $97 \%$. An increase in accuracy is also achieved as body proportions deviate from the group average, suggesting that skeletal uniqueness provides differentiation. These results highlight the potential of pose-based biometric identification, particularly in scenarios where facial data is unavailable or unreliable.