Computing Realistic Images for Audience Interaction in Projection-Based Multi-view Display System

Stereoscopic multi-view system can provide multiple users with individual views of a virtual environment. But the existed systems can only present n (n≤6 currently) images, limited by its ratio to the refresh rate of display set, for users on a shared display set. To support more users, we propose two methods to cluster the users into k (k≤n) groups, so that users can perceive different stereoscopic images in different regions and the same image in the same region. The first method subdivides the virtual space into k=n fixed regions by Centroidal Voronoi Diagram (CVD) in the preprocessing step, where the centroid positions of CVD regions are used as viewpoints to render corresponding images. It runs fast because it only need to trace the positions of users and compute in which CVD regions they are locating in the running time, and the viewpoints can stay stabilized at some time. But it may lead to different views for users who are located in different regions while close to each other. The other method is designed to dynamically divide the virtual space into k≤n regions by k-means method in the running time, where the centroid positions of cluster blocks are used as viewpoints to render corresponding images. It can compute more effective viewpoints and more realistic images. As viewpoints are real-time computed according to the variable spatial-temporal information of users, the number of rendering images can be dynamic during the process, i.e., It is possible to be equal to or less than the maximum number of views the system can display, depending on the distribution of positions of moving users.