This work proposes a simple and effective strategy for photon clustering, called regional photon clustering, to accelerate the global illumination updating of dynamic scenes. This method can cluster various numbers of photons adaptively. The scenes in adjacent frames are similar, and so most photons have the same moving paths in adjacent frames. Therefore, most photons do not need to be shot again in the next frame. Regional photon clustering employs a quadtree structure to track the trajectory of dynamic objects to reduce the number of re-shot photons. Experimental results show that regional photon clustering reduces the number of re-shot photons, and accelerates the estimation of global illumination in dynamic scenes