Verification of the general position assumption in the practice of stereovision

In feature-based stereo analysis the use of restrictive knowledge-based assumptions constrains effectively the processes of interpreting and matching image features. Among these assumptions the general position assumption plays a significant role and is employed in many approaches. But working with complex scenes shows that this assumption is quite often violated for some parts of the scene. The paper proposes an active stereovision approach which completes automatically the reconstruction of a given scene and avoids thereby special views and errors in the interpretation and matching of image features