This study examines AI-based pose estimation algorithms for touchless user interfaces, focusing on gesture-based control through diverse input modalities and user postures. We evaluate state-of-the-art deep learning models for tracking hand, head, and full-body motion, emphasizing their accuracy and responsiveness in dynamic gesture recognition. The study analyzed several AI-based human pose estimation models, including MediaPipe, OpenPose COCO, MoveNet Lightning, and Thunder. We assess gestures of varying complexity performed in two scenarios: users seated at a computer with limited upper body visibility, and users standing with full body visibility. A user study objectively compares the methods under consistent experimental conditions. Results indicate that pose estimation models perform best with full-body input, while head- and hand-based control present more challenges. Although hand gestures are natural for pointing, occlusions reduce tracking reliability. Head-based control, while technically feasible, may feel less intuitive for users.