Image Processing is fundamental for any camerabased vision system. In order to automate the prototyping process of image processing solutions to some extend, we propose a holistic, adaptive approach that comprises concepts for specification, composition, recommendation, execution, and rating of image processing functionality. The fundamental idea is to realize image processing applications according to Service-oriented Computing design principles. That is, distinct image processing functionality is encapsulated in terms of stateless services. Services are then used as building blocks for more complex image processing functionality. To automatically compose complex image processing functionality, our proposed approach incorporates a flexible, Artificial Intelligence planning-based forward search algorithm. Decision-making between alternative composition steps is supported by a learning recommendation system, which keeps track of valid composition steps by automatically constructing a composition grammar. In addition, it adapts to solutions of high quality by means of feedback-based Reinforcement Learning techniques. A concrete use case serves as proof of concept and demonstrates the feasibility of our holistic, adaptive approach.

Electronic systems, like they are embedded in road vehicles, have to be compliant to functional safety standards like ISO 26262 [1], which limit the impacts of malfunctions for safety critical systems. ISO 26262, for instance, defines different safety levels for road vehicles, which require different means and measures for a safety compliant system and its development process like risk analysis and fault effect simulation. For fault effect simulation it is important to investigate the impact of physical and hardware related effects to the correct function of a system. This article first studies code and model mutations for fault injection in the context of fault effect simulation through different system abstraction levels. It demonstrates how high level mutations correlate to bit flips of software binaries by examples from the TriCore™ instruction set and finally presents a virtual platform based implementation for automated injection of bit flip based mutations into software binaries. Experimental results demonstrate the efficiency of the implemented approach.

Automatically composing service-based software solutions is a challenging task. Considering context information during this service composition process is even more challenging. In domains such as image processing, however, context-sensitivity is inherent and cannot be ignored when developing techniques for automatic service composition. Formal approaches tend to create ambiguous solutions, whenever the expressive power of the applied formalism is limited. For example, services may have the same formal specification, although their actual functionality depends on the concrete context. In order to satisfy individual user requests while providing data-dependent functionality, formal approaches have to be extended. We propose to incorporate Reinforcement Learning techniques and combine them with planning based composition approaches. While planning ensures formally correct solutions, learning enables the composition process to resolve ambiguity by implicitly considering context information. Preliminary results show that our combined approach adapts to a static context while still satisfying formally specified requirements.

Automatic service composition is still a challenging task. It is even more challenging when dealing with a dynamic market of services for end users. New services may enter the market while other services are completely removed. Furthermore, end users are typically no experts in the domain in which they formulate a request. As a consequence, ambiguous user requests will inevitably emerge and have to be taken into account. To meet these challenges, we propose a new approach that combines automatic service composition with adaptive service recommendation. A best first backward search algorithm produces solutions that are functional correct with respect to user requests. An adaptive recommendation system supports the search algorithm in decision-making. Reinforcement Learning techniques enable the system to adjust its recommendation strategy over time based on user ratings. The integrated approach is described on a conceptional level and demonstrated by means of an illustrative example from the image processing domain.

After large accidents or natural disasters a fast and efficient retrieval of injured is required. With recent advances, unmanned aerial vehicles (UAVs) are more and more a viable choice under such circumstances. The problem of coordination arises with the number of employed UAVs as well as proper task allocation among possibly heterogeneous UAVs. In this work a hybrid approach for UAV coordination and task allocation is presented. It covers the exploration of unknown terrains as well as role-based, goal-oriented path planning and simultaneous task allocation. The approach uses a role system to support prioritized task assignment and an informed search algorithm to find consistent trajectories to known goal positions. The system is suited for highly dynamic environments, requiring frequent recalculations.

Although the combination of reinforcement learning and imitation has been already considered in recent research, it always revolved around fixed settings where demonstrator and imitator are fixed and the imitation process is a well-defined period of time. What is missing is the investigation of approaches that also work in scenarios where imitation is only sporadically possible. This means that in a multi-robot scenario a robot is now allowed to interrupt another robot by asking to repeat certain actions, but can only observe and integrate information bits delivered occasionally. In this paper we present how that can be done in continuous and noisy environment within an SMDP context.

This paper presents the fuzzy video based emotion recognition system VISBER, that allows to analyze facial expressions in video sequences. In order to process images in real-time a tracking mechanism is used for face localization. The fuzzy classification itself analyzes the deformation of a face separately in each image. In contrast to most existing approaches, also blended emotions with varying intensities as proposed by psychologists can be handled. For this purpose we propose a fuzzy emotion model which is generally applicable for also for other emotion recognition solutions. Furthermore, VISBER supports the automatic adaptation to the characteristics of individual human faces by a short training phase that can be done before the emotion recognition starts.

Emotion recognition and adequate reactions are a crucial part of human communication and hence should also be considered for interactions between humans and robots. In this paper we present the robot head MEXI which is able to recognize emotions of its human counterpart from a video sequence using a fuzzy rule based approach. It reacts on these perceptions in an emotional way. Therefor MEXI maintains an internal state made up of (artificial) emotions and drives. This internal state is used to evaluate its perceptions and action alternatives and controls its behavior on the basis of this evaluation. This is a major difference between MEXI and usual goal based agents that rely on a world model to control and plan their actions. For MEXI the behavior based programming paradigm originally developed by Arkin for robot navigation was extended to support a multidimensional control architecture based on emotions and drives.

This paper presents the robot head MEXI which is able to communicate to humans in an emotional way. MEXI recognizes emotions of its human counterpart from the prososdy of his or her natural speech using a fuzzy rule based approach. MEXI reacts on its perceptions by showing artificial emotions in its facial expressions and in the prosody of its synthesized natural speech. MEXI does not rely on a world model to control and plan its actions like usual goal based agents. Instead MEXI uses its internal state consisting of emotions and drives to evaluate its perceptions and action alternatives and controls its behavior on the basis of this evaluation. For MEXI, the behavior based programming paradigm originally developed by Arkin for robot navigation was extended to support a multidimensional control architecture based on emotions and drives

Due to increasing miniaturization and decreasing prizes of cameras more and more mobile devices like PDAs or smartphones are equipped with a camera. Due to this fact, mobile face recognition will gain popularity in identifying persons e.g. in order to prevent unauthorized use or access to data and equipment. In this paper, the feature based face localization and recognition system FaceScry is presented. In spite of the limited resources available on mobile devices, FaceScry is able to localize an arbitrary number of faces with different sizes in images taken under varying illumination conditions in real-time. Also face recognition is size invariant due to the selected set of features, which mainly consists of angles and cross ratios. Since it stores reference face data for recognition as feature vectors and not as huge image data, FaceScry also allows for keeping a reasonable personal face data base for recognizing a set of persons on the smartphone

Distributed real-time computing is one of the enabling technologies in establishing a team of autonomous robots. A team of soccer playing robots, Paderkicker, contains many computing nodes handling environment perception, artificial intelligence, team coordination, autonomous actuation, and interactions with external entities such as a remote server. Robots must act with varying degrees of autonomy in real-time. The time-triggered message-triggered object (TMO) programming scheme offers a high-level high-precision approach to programming of distributed real-time computing operations. The TMO programmer can apply deadlines to method executions and result-arrivals in a straightforward way. He is thus relieved from dealing with low-level resource allocation details and can show considerably increased productivity. This paper describes a TMO structuring approach for constructing teams of autonomous robots. A TMO network architecture is described for the Paderkicker team.

This paper describes the emotion recognition from natural speech as realized for the robot head MEXI. We use a fuzzy logic approach for analysis of prosody in natural speech. Since MEXI often communicates with well known persons but also with unknown humans, for instance at exhibitions, we realized a speaker dependent mode as well as a speaker independent mode in our prosody based emotion recognition. A key point of our approach is that it automatically selects the most significant features from a set of twenty analyzed features based on a training database of speech samples. This is important according to our results, since the set of significant features differs considerably between the distinguished emotions. With our approach we reach average recognition rates of 84% in speaker dependent mode and 60% in speaker independent mode.

In this paper an object recognition and tracking approach for the mobile, marker-less and PDA-based augmented reality system AR-PDA is described. For object recognition and localization 2D features are extracted from images and compared with a priori known 3D models. The system consists of a 2D graph matching, 3D hypothesis generation and validation and an additional texture based validation step.

This paper presents VisiTrack, a novel approach for video based incremental tracking in real time. The major objectives in the development of VisiTrack was to design or select algorithms that are well suited for embedded real time computation. We had a special focus on latency reduction and storage minimization since the algorithms should run on mobile devices like PDAs with the appropriate extension, i.e. mainly a camera, in real time. The image analysis, position localization and error minimization of VisiTrack are explained in detail. For the image analysis algorithms of VisiTrack two major principles, linear processing and incremental concurrency, were defined and applied in order to fulfill real time constraints and reduce system latency. In order to meet real time requirements the position localization is based on a linear transformation between 3D world coordinates and 2D image coordinates. Afterwards a real time error minimization based on the Kalman filter is applied successfully. VisiTrack is applied in the Augmented Reality domain and in the field of robotic self localization.

This paper presents a model for real-time computer vision applications based on the synchronous dataflow (SDF) methodology. To accomplish the needs of computer vision applications (mainly access to pixel neighborhood and access to previous video frames) a new buffer concept called Structured Buffers replaces the FIFO buffers used in the SDF model. The model minimizes the system's latency and memory consumption and allows an execution time analysis which is an essential prerequisite for real-time applications. Finally a line-based edge vectorization is presented as an example of the usage of the presented model.

An approach for the synthesis of highly complex controllers is presented. As source description a concurrent algorithm is used. It is assumed that usually such a description is organized in a modular manner or can be partitioned into modules by the synthesis algorithm. The synthesis algorithm produces cooperating clocked FSMs with an asynchronous global communication mechanism. Modified Petri nets are used for the specification of the concurrent algorithm to be implemented and as implementation method. In this paper we describe how the FSM modules are identified by the implementation algorithm and mapped onto standard realizations. Furthermore, our implementation method of the asynchronous inter-module communication is presented.