Finding persons in a video or recognizing text of handwritten documents are current research topics in the field of computer vision.

In practice, finding groups of pedestrians pose great demands on the detection methods. Similarly, the detection of flames or smoke is an interesting problem: partial transparence, and the continuous change of form makes it difficult to describe these 'objects'.

In this seminar we focus on the following methods to cope with these problems:

• Text recognition: Countless handwritten documents from the Early Middle Ages lie hidden in archives. The large amount of different writing styles raises the demands for an automatic text recognition. Due to a cooperation with the history department we have access to documents from the eigth to the twelth century.
• Detection of hunmans: The detection of humans finds an application in the fields of robotics, driver assistance systems and surveillance. Hereby, the main problem of detection lies in the large variety of poses, differences in appearance, and the different sizes of humans.
• Dynamic textures: How do you detect objects without a solid form? Smoke, waves or flames change their appearance from frame to frame. However, their movement exhibits specific patterns which are detectable.

Every participant will develop a prototype, based on current publications. The performed experiments for the detection or recognition task are then presented to the group and summarized in a short report.