Efficient visibility masking for crowd rendering with shadows

Virtual real-time applications such as games and simulations frequently require rendering of large number of agents. Various GPU accelerated eye space culling and level of detail management techniques are being used to achieve efficient rendering of massive virtual crowds. However effective light space rendering is important for shadow mapping. Common techniques use partitioned shadow maps for better quality to utilize detail management which introduces the problem of culling and rendering the crowd multiple times. Similar problem also arises when using local point and spot lights with shadows. One of the previous works offers using the light's radius and a spatial hierarchy to cull unnecessary agents while rendering the shadow maps of the local lights[Ryder abd Day 2000]. But efficient shadow caster culling becomes crucial to reduce overhead of rendering massive crowd geometry multiple times.