Real-time Human Pose and Shape Estimation for Virtual Try-On Using a Single Commodity Depth Camera.

Area of Science:

• Computer Vision
• Computer Graphics
• Human-Computer Interaction

Background:

• Virtual try-on technology aims to enhance online shopping experiences.
• Accurate 3D body reconstruction from monocular depth cameras presents significant challenges, including data occlusion and pose ambiguity.

Purpose of the Study:

• To develop a real-time virtual try-on system using a single depth camera.
• To address challenges of incomplete body data and pose ambiguities in monocular 3D reconstruction.

Main Methods:

• A novel real-time template-based approach for joint pose tracking and shape adaptation.
• Utilizing automatic shape adaptation and new constraints in pose tracking to handle missing data.
• Driving realistic cloth simulation with the estimated 3D body model for virtual try-on.

Main Results:

• The system successfully estimates user pose and shape in real-time from a single depth camera.
• Demonstrated effectiveness in overlaying synthesized clothing onto the user's image with realistic simulation.
• Successfully addressed challenges posed by monocular vision, such as partial body capture.

Conclusions:

• The developed system offers a viable solution for virtual try-on applications.
• The novel approach enhances the accuracy and realism of virtual clothing simulation.
• This technology has the potential to improve online retail and customer engagement.