Real Time Scale and Rotation Independent Object Detection

Abstract

Images and videos can include different kind of semantic content, which is an important resource for erous application areas when certain information is extracted from them. For instance, in the area of media monitoring advertisement statistics are collected from visual media and commercially offered. At the time, the automatic extraction of this information is technically restricted and thus the needed information is annotated manually while most information is not used at all. Object recognition systems are a promising approach to achieve a further automation of this task by automatic annotation of the position, size and pose of recognized objects in images and videos. The goal of this thesis is the development of such an object recognition system for the use in automation monitoring of the object. Object recognition is a difficult problem because the same object could appear in many different ways in real-world images. An object could be occluded and a photo of this object could be taken from different views or under changing illumination conditions, so that the object appears with an arbitrary size, anywhere in the image. In the last years, many object recognition systems were built, which are well suited to overcome several of these difficulties by the use of local features. Such local features are extracted in a large number from all parts of an image and they describe the visual content of these image regions. Our object recognition system uses local features from the different algorithms, which seemed to fulfil our requirements best in a previously performed state of the art analysis. Therefore we have approached an object recognition algorithm to recognize angle and translation of the object.