In this paper, an algorithm is presented to generate visual hulls of 3D objects. The algorithm fully utilizes the advantages of ADFs (adaptively sampled distance fields) in approximation precision and memory requirement. Different from the traditional voxel interpolation methods, the algorithm takes ADFs as the data structure to describe the visual hulls to simplify the interpolation process and improve the algorithm efficiency. Finally, the enhanced SurfaceNets algorithm is employed to triangulate ADFs in order to obtanin the topologically consistent models.The experimental results show that the algorithm can rapidly generate complex models with high precision.