Final Technical Report 5 2 Description of Complex Objects from Multiple Range Images Using an Inflating Balloon Model Yang Chen and Gérard Medioni We address the problem of constructing a complete surface model of an object us- ing a set of registered range images. The construction of the surface description is car- ried out on the set of registered range images. Our approach is based on a dynamic balloon model represented by a triangulated mesh. The vertices in the mesh are linked to their neighboring vertices through springs to simulate the surface tension, and to keep the shell smooth. Unlike other dynamic models proposed by previous re- searchers, our balloon model is driven only by an applied inflation force towards the object surface from inside of the object, until the mesh elements reach the object sur- face. The system includes an adaptive local triangle mesh subdivision scheme that re- sults in an evenly distributed mesh. Since our approach is not based on global minimization, it can handle complex, non-star-shaped objects without relying on a carefully selected initial state or encountering local minimum problem. It also allows us to adapt the mesh surface to changes in local surface shapes and to handle holes present in the input data through adjusting certain system parameters adaptively. We present results on simple as well as complex, non-star-shaped objects from real range images. 2.1 Introduction The task of surface description using 3-D input can be described as finding a fit of a chosen representation (model surface) to the input data. This process can be for- malized in a number of ways involving the minimization of a system functional that explicitly or implicitly represents the fit of the model to the input data. Another very important aspect of such a system is to construct a mapping or correspondence be- tween the surface of an object and the structure of the model. This mapping exists be- cause the surface of the model and the surface of the object are topologically equivalent, considering genus zero type of objects. Therefore there exists a one-to-one mapping between the model structures and the object surface elements. Previous re- searchers have studied such mappings in a variety of ways using different represen- tation schemes and model fitting methods. Examples of these approaches include the dynamic system using energy minimization in [4] and the dynamic mesh in [12] and [13]. The drawbacks of these approaches is that they must rely on an initial guess of the model structure which is relatively close to the shape of the object. The reason is that, in the absence of mapping or correspondence information, some other approxi-