Feature Detection and Matching. Computer Vision

Abstract or Introduction

This thesis deals with the eminent as well as demanding subfield of computer vision for the implementation of image processing and image analysis processes. Explicitly meant here is the research field of "Feature Detection and Matching". In concrete terms, this field of research comprises numerous proven and tested detection algorithms for the calculation of abstract image information as well as for local decision making at image points for feature recognition.
The tangible application of this form of technology takes place in many sub-competencies of computer vision. These include the joining of image mosaics, image and video stabilization, and the recognition and/or match analysis of image object instances.

There are two main principles of feature recognition: Feature Matching, describes the viewing and recognition of all features within an image object and the assignment of these based on their local features, feature Tracking, describes the analysis of image features with local search techniques, such as correlation, to find and track image features.

The main goal of the scientific work requires the definition and explanation of essential features of the applied methodologies in a manageable abstract form. In doing so, numerous recognized fields of computer vision and applied mathematics are included for the purpose of argumentation and proof. Of elementary importance is the constant attempt to show a smooth transition between concepts that are particularly theoretically mathematically based and the actual practical application and use. At this point, numerous specially developed software elements in Python are used with the OpenCV library to illustrate the practical part.

The main source of this scientific work is the extensive book "Computer Vision: Algorithms and Applications" by Richard Szeliski as well as all other scientific papers of the experts listed in his book.
The completion of this thesis is accompanied by a self-developed software, which uses the discussed methods for feature detection and matching to perform image processing and image analysis processes.