Recovery of superquadric primitives from a range image using GA
Hideki Tanahashi, Noriyasu Murakami, Kazuhiko Yamamoto · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1997
This paper describes a method for recovering volumetric shape descriptions of 3D objects from range image. The models of shape primitives are superquadrics. The model parameters defined the shape are obtained by using genetic algorithm. The fitting energy evaluated the difference in both depth form the object center and the distribute of the normal vector between input range image and the model surfaces. When the next generation is made the model parameters are independent each other, so the crossover is applied each parameter independently. In order to overcome the local minima problem, we employ a mutation strategy and a elitist preserving strategy. After the model parameters are obtained, the remained parameters are set. We applied the proposed method to recover nondeformed superellipsoids from synthetic range image.