Superquadrics model recognition from stereo image

This paper presents the process of recognition of the three-dimensional object and calculation of similarity. The object is recovered from stereo image and superquadrics function is used as the model. The superquadrics is a parametric and volumetric model. It is represented by an expanded ellipse function. And genetic algorithm is used for recovering of an object shape. A fitness function of genetic algorithm is defined using texture features of object on the stereo image. The features are shared area and difference of intensity among the left image and the right image. Parameters of superquadrics which are replaced by parameters of genetic algorithm are optimized and the superquadrics form fits to a solid body object. In the case of multiple body objects, the object is divided on the image, and each solid body is recovered. The objects used in the experiment are a cube, a cylinder and an elliptic column. The algorithm is useful for the recovery of the object under stereopsis. After recovery, the similarity between the recovered object and the model in the computer is calculated.