Fast 3D modeling from images

Abstract

Visual hull of object has been widely applied in many fields in computer vision. The methods computing visual hulls are mainly classified into two categories: surface-based approaches and volume-based approaches. Surface-based approaches are precise and lack robustness while volume-based approaches are robust but inaccurate and work slowly. However, the speeds of both methods cannot satisfy the demand of real-time application. The paper proposes a novel method to compute visual hull rapidly. To this aim, the method is based on viewing lines. First, the viewing lines are computed from all the contours of the object from different viewpoints. Secondly, the viewing lines are confined within limits and sampled into some dispersive points. Thirdly, a sequence of images of the object is used to exclude points which locate outside the surface of the visual hull. Finally, a water-tight surface of visual hull is extracted from the remnant points. Experiments with real data are conducted to test the rapidity of the proposed method.