Procedural Generation of 3D Canyons

This paper presents a non-assisted method for procedural generation of 3D canyon scenes. Our approach combines techniques of computer graphics, computer vision, image processing and graph search algorithms. Our main contribution is a generation approach that uses noise-generated height maps that are carefully transformed and manipulated by data clustering (through the Mean Shift algorithm) and image operations in order to mimic the observed geological features of real canyons. Several parameters are used to guide and constrain the generation of terrains, canyon features, course and shape of rivers, plain areas, soft slope regions, cliffs and plateaus.