During the last decade, symmetry estimation has gained importance in the field of computer graphic and computer vision. It has been successfully used in many applications, including segmentation, shape completion, and object recognition. The aim of this thesis is to estimate the bilateral symmetry of a 3D object represented as a set of 3D points from image sequences. The proposed algorithm to solve this task is based on gathering pairs of 3D points that reflect the bilateral symmetry property of the 3D object of interest.
Two 3D points built a symmetric pair when their corresponding 2D image points in each image frame have the same descriptive features, i.e, the image patches associated to the descriptive features represent two mirrored regions of the 3D object. The gathered set of pairs is then used to find the optimal reflective symmetry transformation such that this set is reflectively symmetric with respect to it. The experiments made on both synthetic data, and real world scene data have shown good results. Particularly, in the case of synthetic data, where the amount of false symmetric pairs is kept small, the algorithm is able to estimate the symmetry plane with a high precision. The conclusion drawn from this thesis is, if the symmetry matching task between the 3D points is reliable,then the proposed algorithm provides results with a high accuracy. Therefore, the design of the algorithm makes it dependent on the quality and robustness of the feature detection and feature matching methods.