Directional Voting with Geometric Hashing for Three Dimensional Object Recognition
Shun’ichi Kaneko, Masahiro Shibata, Tsunenori Honda · 1994
the combinations of triplets of basis points from the set of feature points and then the HT can include normalized A new method for recognition of three dimensional information of feature distribution on the image plane. In objects by use of a monocular image is proposed. It the recognition phase, a set of feature points from the utilizes sample monocular images of objects as aspect image of an object is transformed with respect to a triplet models and a geometrical transformation and a directional of basis points and then according to coordinates of voting procedure are essential for robust recognition transformed points the model which gets the maximum against feature defects such as aspect changes or occlusion. voting score is selected as the corresponding model. With the method, the sensitivity of recognition of the To apply the geometric hashing technique to three correct model can be kept high while the amount of models dimensional (3D) objects, 3D information processing has (a dictionary) is reduced. Experimental results with real been necessary to construct a HT, which involves a method objects show an effectiveness of the proposed method. based on 3D shape data (CAD data) and quadruplets of basis points[4] or another method based on two (stereoscopic) images and a corresponding procedure [12].