For The Depth Of The Image Block Object Parts Modeling And Segmentation

Model-based objects segmentation is a very crucial step in the processing of range image modeling and recognition. In a multi-object scene, because the objects are covered each other, sampling data is lost partly. This paper researched and designed the method of modeling objects with data lost partly using superquadric rim and the method of range images region segmentation based on the edge growing, and implemented the model-based modeling and segmentation of covered objects.First of all, this paper used superquadric projecting rim to model objects with data lost partly. Through analyzing the rim fitting principle of superquadric, an initial objective function for superquadric rim fitting was set up. Simultaneously, the function was optimized by using equal-distance sampling and projecting rim. By designing and implementing the rim fitting algorithm of superquadric, the superquadric parameters corresponding to the object were obtained, and the original model with data lost partly was recovered successfully. Secondly, for the case which the objects are covered each other, the objects are modeled and segmented by using the region growing method based on edge and region restricting. Under analyzing the fundamental principle of the region growing and the basic process of range images segmentation, researching and designing the region growing method based on edge and region restricting, the region information and the edge information in the range images were merged each other, then the growing rule was improved. Finally, by designing and implementing the modeling and segmentation algorithm of model-based object for the range images, we got the superquadric parameters of model-based objects which had been segmented. The simulation results showed that the rim fitting method of superquadric implemented modeling object with data lost partly, and the region growing method based on edge and region restricting implemented modeling and segmentation of covered objects. These methods laid a foundation for model-based object recognition.