CHAINING APPROACH FOR 3-D OBJECT VOLUME RECONSTRUCTION

摘要

From a gray level 2-D picture, objects recognition is possible within a scene. Beyond this major step, robotics, CAD and image synthesis, require, in addition to the vision, complementary data to satisfy information need, such as: 1. precise knowledge of objects surface and volume, in case robots can handle, gather and control their quality; 2. robot environment knowledge, enables to work and avoid obstacles while moving; 3. objects shape knowledge for CAD to ensure modelling, analysis and planes execution; 4. objects apparent points knowledge, useful for image synthesis and graphs. In this paper, a new method for 3-D object volume construction is proposed, using the chaining approach. The constructed 3-D object can be viewed on the screen using perspective or parallel projections under any viewpoint angles θ and φ A one-pixel precision is reached for recovered objects.