Robust curve approximation for 3-D model-based vision

The recognition of curved objects is a challenging problem in computer vision. The identification of the shape should be possible in the presence of occlusion and noise. This implies that shape analysis ought to be local, enabling partial matching techniques in recognition phase. A match between an object and the model is carried out on intermediate representation level quantities that are referred as features. The best match occurs when the hypothetical object model configuration represents the scene world as accurate as possible. The accuracy can be measured quantitatively by the matching errors.