The Research Of Real-time Shadow Algorithms In Dynamic Scenes

Shadow rendering plays a very important role in the generation of realistic images.The shadow mapping algorithm can render shadow efficiently and scale well with the scene complexity.What's more,it is compatible with modern graphics hardware,so it is widely used in real-time applications.In this paper,we consider the soft shadow algorithm based on shadow maps.These algorithms use the filtered shadow maps to achieve the purpose of anti-aliasing and rendering soft shadows.First,we analyze the theory of previous algorithms,and give out their advantages and disadvantages.These algorithms are only applied to the scene which has a one-directional light source.Concentric Spherical Representation for Omnidirectional Soft Shadow(OSM)can render soft shadows in scenes which have an omnidirectional light source,but it inherits the Convolution Soft Shadow Maps(CSSM)'s shortcomings,which requires lots of memory to save the base images,and produces light bleeding in certain circumstances.Based on Moment Soft Shadow Maps(MSSM),we propose an omni-directional soft shadow mapping algorithm,which only needs a 4-channel cube shadow map to quickly generate high-quality omni-directional soft shadows.In the experiments,we use the M ip map technology and the summed area table technology(SAT)to filter the cube shadow map.For the aliasing problem caused by the typical Mipmap,we introduce a multi-sampling scheme to improve the filtering effects.The effect of the soft shadows is similar to the shadow generated by the algorithm based on SAT,but has a faster rendering speed.When using the Percentage Closer Soft Shadows(PCSS)framework for anti-aliasing and soft shadow generation,a large depth bias is required to eliminate the over shadow phenomenon.However,the use of a large depth bias will cause light bleeding on the contact point of the receiver and blocker.To solve this problem,this thesis proposes a percentage closer filtering algorithm based on angle calculation.The algorithm can eliminate the wrong shadow rate calculated by the inappropriate use of filter kernel,so only a small depth bias is needed to deal with the self-shadowing caused by the accuracy problem of the depth map.