Model-Based Keypoint Filtering - A Data Mining Approach

Christopher Kuhn
(19.07.2018, 13:15)
Raum: 0406
Feature detection and feature description is of great importance in most computer vision applications such as object classification, image stitching and self-localization. There exists a variety of algorithms for visual feature extraction. In this thesis machine learning algorithms were evaluated (e.g. regressors of Deep Networks, Logistic Regression, Ensemble Methods and Meta Classifiers) to rank the importance of keypoint properties. Keypoints (KP) are the output feature detection algorithms and have variables such as KP-strength, KP-location, KP-size or KP-scale. After finding appropriate training data machine learning algorithms were employed to automatically rank the properties of keypoints to generate the overall score of the importance of the keypoint. This technique helps discarding meaningless keypoints to save further processing.
Betreuer: Martin Oelsch