Ralf Stemmer

Evaluation of Visual Attention Models for Robots

This paper presents a new approach for providing visual attention on robot vision systems. Compared to other approaches our method is very fast as it processes regions rather than individual pixels. The proposed method first builds a list of regions by applying a shade and shadow tolerant segmentation step. The features of these regions are computed using their convex hulls in order to simplify and accelerate the processing. Feature values are stored within the records of respective regions instead of constructing a master map of attention. Then an algorithmic method is applied for finding the focus of attention in contrast to mathematical approaches used by existing models. Experiments conducted on simulated and real image data have not only demonstrated the validity of the proposed approach but have also led to the establishment of a comprehensive robotic vision system.

Vision-based control of an autonomous disassembly station

Abstract In the EU about 10 millions of used cars have to be wrecked per year. So the disposal of these cars is of increasing importance in the next years. In a disassembly process for used cars one has to cope with the objects in an undefined state in contrast to the car production. For this purpose the use of vision-based control is motivated and a vision guided robotic system for an autonomous disassembly process is presented in this paper. An active stereo camera system is used as vision sensor. A combination of gray value and contour-based object recognition and a position measurement approach with implicit detection of occlusions is described. A system for the autonomous disassembly of wheels was successfully tested and shown in a final presentation. In this demonstrator a stereo vision system, a force–torque sensor, a task planning module, and a special unscrewing tool were integrated.

Comprehensive data acquisition for a telepresence application

This contribution presents the newly developed platform for telesensory robotic applications TSR. An operators PC, which represents the man machine interface, enables an operator to interact with the radio-controlled robot. To provide an impression as close as possible to reality, the transmission and presentation of visual information is of special interest. Therefore, the video streams which are grabbed from a stereo camera system are presented to the operator on a head mounted display such that a three dimensional impression of the robots environment is achieved. To increase the operators immersion, additional information of further sensors are presented in the HMD or transformed into forces on haptic input devices. Additionally, the capability of data processing on board TSR allows a (partly) autonomous operating status and the assistance of the operator.

Region-based Depth Feature Map for Visual Attention in Autonomous Mobile Systems

In this contribution we present a fast region based approach for distance estimation within a bottom-up visual attention model. The feature “depth” is utilized to focus the attention on objects that are closer to the mobile robot in order to accelerate the overall process of visual attention. Due to the fact that in the stereo algorithm we only search for corresponding segments determined by color segmentation, we provide a fast method for usage in mobile systems. Although we do not achieve very dense depth maps, the accuracy is sufficient for collision avoidance and the integration of the obtained depth map into the atte

An active object recognition system for disassembly tasks

Recycling of used products is of increasing importance in a society that becomes aware of its responsibility for the environment. Thus, the automatic disassembly in recycling processes will be of increasing importance within the next decade. In this paper, we motivate the use of vision-based control of robot manipulators in an autonomous disassembly station. For this purpose, we present an active object recognition and measurement system based on a stereo head with four mechanical and six optical degrees of freedom. The system has been successfully tested for the task of automatically unbolting wheels from unknown used cars.

Dreidimensionale Lagebestimmung zu demontierender Objekte mit einem aktiven Stereokamerasystem

In diesem Beitrag wird ein aktives Stereokamerasystem zur visuellen Steuerung eines Demontageroboters beschrieben. Durch ein neuentwickeltes Stereoverfahren wird im Sinne einer Grob- Fein-Strategie eine erste grobe Tiefenschatzung durchgefuhrt und mogliche Hindernisse im Arbeitsbereich des Roboters lokalisiert. Parallel dazu werden die zu demontierenden Objekte erkannt und nach der Erkennung durch das aktive Sehsystem fixiert. Durch Erhohung des Zoomfaktors und Filterung der Bilder in hoher Auflosung werden die dreidimensionale Position und Lage der Demontageobjekte im Raum genauer ermittelt. Die durch das Stereoverfahren errechneten Weltkoordinaten werden in Roboterkoordinaten transformiert, um eine Demontage des Objektes durch den Roboter zu ermoglichen. Die Positions- und Lagebestimmung der Objekte ist fur eine Demontage von Radern hinreichend genau.