3D object recognition and pose estimation using feature descriptor regression in a Bayes' framework
Sergi Segura Morros · 2012
[ANGLES] In this thesis, we have tried to find a suitable method to solve typical applications of pose recognition for cars, face, or facades. We have looked for efficient algorithms that allow us to solve the problem without the need of reconstructing a 3D model of the object and therefore, with a much lower computational load. The method mimics the first steps of a 3D reconstruction, where we need to take pictures of the object at different orientations, but, instead of building a computationally complex 3D model of the object, we use the information extracted in the feature descriptors of each image to estimate the feature appearance at unknown poses. We can take advantage of the fact that descriptors change their values when a change in the orientation of the object occurs, and predict the values at orientations for which the ground truth information is not available. The method is separated in two parts, the Off-line and the On-line Stage. In the Off-line Stage, we take pictures in a few known poses of the object to recognize, and we establish a track for each feature along the available images. For each feature track, we build a regression function that will estimate the value of the feature at unavailable poses. In the On-line Stage, a test image is input to the system. We extract its features and compare them with the features of the available training poses to establish correspondences. Once this matching is done, and following the principles on which SIFT features are matched, we compute the Euclidean distance between each feature in the track and the test image to find the most similar one. In order to achieve a more accurate result, we estimate the value of the feature at the poses that are not available by applying the regression function at those orientations. The pose estimation is conceived as an optimization problem as we have to minimize the error function given by the distance between the estimated descriptor and the current one. As the error function presents various local minima (the error function is not perfectly concave), we divide it into windows and then choose the global minimum among them, retrieving in this way the correct pose of the test image. The other main reason to divide the domain in sub-intervals is to maximize the number of tracks used. By embedding the minimization inside a Bayesian framework, we can estimate the probability of the actual pose given the feature descriptors of the test image.