To introduce mobile robots into a home environment, ways for interaction with this environment need to be provided. One attempt is to teach robots to recognize objects with the help of learning algorithms. To train these algorithms to efficiently recognize objects like chairs, couches or tables, a large database of images containing these objects is required.
The existence of certain objects within an image can be expressed with image annotation.
The term annotation generally denotes an extra information added to a document.
Therefore image annotation is additional information superimposed on an image.
Images can be annotated in a variety of ways. One way would be to use a graphics manipulation program to add a named layer on top of the object. The drawback of such approach is that loading and processing of the individual images takes too long time. Thus, a tool for efficient annotation of large image databases is necessary.
Nowadays many tools exist for applying annotation to images and creating annotation databases. Unfortunately, most of those tools weren't made available outside the research groups where they were developed. Tools that are available are often designed to solve a specific problem and are not easily applicable for more general tasks.
In this thesis we present Ultimate Annotation, a new application for efficient, semiautomatic annotation of image data. The way we attempt to annotate images is through the use of polygons. Polygons allow to tightly surround the object's silhouette, and efficiently represent annotation. We make use of a line approximation procedure and a segmentation procedure to achieve a semi-automatic behavior. We provide means for quick browsing through a large number of images. The user interface is designed to support quick access to application functionalities, mainly through mouse and keyboard.
The project is open and has a modular composition which allows other groups to contribute and add extensions. All these features combined help to efficiently build large databases of annotated images.