| Rendering large data sets is still a challenge in computer graphics. Current trends lead to a greater number of smaller primitives, to provide more detail. One alternative that scales well is to represent scenes as sets of points instead of the conventional triangle primitives.; We present a new method for rendering very complex point-sampled scenes in real time. We start with an overview of our rendering system. Then we propose a multiple view visibility sorting order for point-sampled data and prove its correctness. The new visibility algorithm is significantly simpler than previously proposed approaches and results in simple memory access patterns, consequently providing higher performance. We also propose an image reconstruction method from non-uniformly distributed projected point samples. Furthermore, we use parallelization to achieve better rendering speeds.; Finally, we describe the practical considerations and implementation details of our work. We present results of our work and outline possibilities for future work, including potential hardware implementations. |
