Distributed Parallel Photon Mapping Method For Large Scale Scene

At present,ray tracing algorithm is one of the most popular and powerful techniques in image synthesis.It is simple,elegant and easy to implement.Although ray tracing algorithms can simulate real light paths and imaging processes,some aspects of the real world are not well handled.For example,the algorithm can not simulate many cases of global illumination,such as diffuse.Photon mapping is a comprehensive method for a consistent global illumination computation with efficient support for caustics.Moreover,photon mapping can also efficiently calculate the illumination effect in participate media.Photon mapping algorithm can also be used to render scenes containing complex objects,which is not limited to tessellated models.Moreover,photon mapping algorithm can deal with advanced materials based on specular reflection,diffuse reflection and non-diffuse reflection?More importantly,the method is simple to implement and easy to experiment.Therefore,photon mapping algorithm is an improvement of ray tracing algorithm.But for large indoor scenes,memory constraints and rendering time are still difficult to overcome.Researchers have provided a method to rendering hundreds of millions of triangles at interactive frame rate[1],but computing time and memory bottlenecks are still great challenges.For large scenes,many researchers divide the scene into several parts by binary space partition[2],so that the scene can be loaded into multiple nodes,so that the memory bottleneck of scene loading can be solved.Portals are inserted between each partitioned unit to facilitate communication between nodes.After scene partition,sub-scenes are distributed loaded using multiple nodes,after that,every node generates its own photon map to store photons.It can solve the bottleneck problem of photon storage.Moreover,every slave node renders one part of the scene at the same time,and at last,the master node is responsible for integrating rendering results into one picture.In this case,task balance between nodes becomes particularly important.In this paper,we propose a distributed photon mapping algorithm based on space partition with virtual portal for indoor scenes such as museums.Our method divides the scene into parts based on the outer contour polygon ofthe scene and inserts a virtual portal between the divided parts.Each node loads only a part of the scene and generates its own photon map in order to solve the memory bottleneck.In addition,this space partition method can ensure that the area of the divided parts is approximately equal and the sum length of the virtual portals shortest.When the height of the virtual portal is the same,the total surface area of the portal is the smallest.Consequently,the load of nodes is more balanced and the data transmission between nodes is minimal.Experimental results show that our method can achieve higher parallel efficiency than the traditional k-d tree based scene partition method.