Interactive network rendering based on textured depth map re-projection

Abstract

Network Rendering is an important problem in computer graphics and visualization

applications. Reduction of the data needed for transmission over the network can greatly

improve the performance. Many methods, such as mesh simplification, image tiling, level

of detail and user action prediction, have been developed to solve this problem..

In this thesis, a new approach tackling the problem has been explored. A textured depth

map (called Terrain Instance Map, TIM) based on re-projection method is introduced

to render highly detailed terrain models. TIMs are created by a rendering server with

a specified viewpoint from the client. Re-projection of the textured depth map in TIM

can satisfy user requests for rendering in the client. In order to achieve an interactive

frame rate at the client side and to reduce the workload of the server, a client side

cache system is designed to reconstruct in-between screen shots from previously rendered

results sparsely provided by the server. Experimental results show that the proposed

method is effective for network rendering.

By changing the projection plane and splitting the viewport into small blocks, Screen

Instance Map (SIM) is developed for arbitrary models with or without texture. The

implemented network rendering system can satisfy interactive rendering in the client.

Compared with existing methods, the new method needs virtually no additional storage

space for the server and supports a wide range of inputs other than meshes.

Finally, computation of camera frames for the virtual camera control is studied as a part

of the network rendering system. A new Single Reflection method (SRM) to solve the

rotation-minimizing directed frame (RMDF) problem is proposed and it is shown that

the method is more robust and twice as fast than the existing method. Based on the

new method, a simple and effective scheme for camera frames interpolation is devised.

The methods and concepts introduced in this thesis are useful and effective for interactive

network rendering. The TIM and SIM methods can be applied to network rendering for models with or without texture. The SRM method is essential for real-time camera

frame computation and its extension can be applied to user interaction for viewing

virtual environments.