Real-Time Clothing Virtual Display Based on Human Pose Estimation
Wenlu Xiong, Zengbo Xu · Frontiers in artificial intelligence and applications · 2024
This paper explores clothing virtual display technology based on human pose estimation and applies it to the Unity3D platform. This technology combines human pose estimation and virtual character-driven approaches to achieve rapid real-time clothing try-on and display effects. Firstly, key point detection of the human body in input images or videos is performed using human pose estimation techniques. This accurately infers the positions and connectivity relationships of human body keypoints, capturing subtle changes in human pose. Kalman filtering techniques are applied to enhance the stability and accuracy of pose estimation. Finally, we apply virtual character-driven technology to the Unity3D platform, developing a real-time dressing application that incorporates the estimated human pose results into virtual character models, enabling real-time clothing changes. By animating virtual characters, dynamic clothing display effects can be achieved, allowing users to instantly view and experience the effects of different clothing styles in various poses.