The embodiment of the invention provides a shadow volume optimization method based on a KD tree; the method comprises the following steps: extracting sharing edges, located on two sides of a light source, of adjacent surfaces as shared contour edges of the shadow volume; segmenting scene data according to the KD tree, extracting the edge with only one adjacent surface to make shading calculation, and extracting boundaries with no shading as the shadow volume boundary contour edges; building the shadow volume according to the shared contour edges and boundary contour edges; segmenting the shadow volume data according to the KD tree, using the shadow volume to remove a bounding volume, removing shadows, and improving the shadow volume peak processing rate. The shadow volume optimization method can solve the problems that a conventional shadow volume algorithm cannot process a non-closed geometry scene, and uses the CPU end shadow volume cutting works to greatly save data bulk inputted into a GPU, thus reducing the GPU burdens.