Technology

The virtual dressing room consists of a room with several cameras which are able to track the 3D motions of a human body in real time and without the need for special markers. A 3D model of the body is animated and dressed in virtual clothes. Finally, an augmented reality mirror-image is shown back to the user on a large display. This will give the user the impression that he or she actually wears the garment shown on the screen.

One usecase of this project is in fashion stores where a potential customer picks some clothes he or she wants to wear, but unfortunately they do not come in the desired size or color. With the virtual dressing room this situtation is no longer a problem as the customer simply steps in front of the intelligent mirror and instantly sees how the garments look on his or her body.

The customer enters a special dressing room with multiple cameras which record his/her 3D appearance as well as the visible image. With this setup a full 3D model is fitted to the body shape of the customer in realtime.

For high quality body measurement and tracking, we use multiple Point Grey Flea2 cameras as they support automatic synchronization of multiple cameras via the firewire bus.

The user is mirrored using an image-based rendering technique called image-based visual hulls (IBVH). This allows the user to watch himself from arbitrary viewing direction.

Coherent regions in subsequent images are detected and reused to minimize the overall latency of the system.

Virtual clothing can be created from a mesh file and physically animated to match the users shape and current pose. The underlying base geometry from physics simulation and other effects utilize a deformed and skinned template mesh.

IBVH rendering in combination with virtual cloth meshes (left) and reconstructed clothes (right) can be used to realize a virtual dressing application.

This video shows the Open Lab Night 2011 demo of the Narkissos project. The setup consists of a projector, a Microsoft Kinect and a PC. It tracks the user and produces an image of the Daz3D Victoria4 avatar that resembles the user's body pose (Youtube link).