Andreas Klausner

Distributed Multilevel Data Fusion for Networked Embedded Systems

Recently much research has been conducted in visual sensor networks. Compared to traditional sensor networks, vision networks differ in various aspects such as the amount of data to be processed and transmitted, the requirements on quality-of-service, and the level of collaboration among the sensor nodes. This paper deals with sensor fusion on visual sensor networks. We focus here on methods for fusing data from various distributed sensors and present a generic framework for fusion on embedded sensor nodes. This paper extends our previous work on distributed smart cameras and presents our approach toward the transformation of smart cameras into a distributed, embedded multisensor network. Our generic fusion model has been completely implemented on a distributed embedded system. It provides a middleware which supports automatic mapping of our fusion model to the target hardware. This middleware features dynamic reconfiguration to support modification of the fusion application at runtime without loss of sensor data. The feasibility and reusability of the I-SENSE concept is demonstrated with experimental results of two case studies: vehicle classification and bulk good separation. Qualitative and quantitative benefits of multilevel information fusion are outlined in this article.

I-SENSE: Intelligent Embedded Multi-Sensor Fusion

I-SENSE demonstrates the potential of combining the scientific research areas multi-sensor data fusion and pervasive embedded computing. The main idea is to provide a generic architecture which supports a distributed online data fusion on an embedded system. Due to their high onboard processing and communication power our proposed architecture is designed to perform sophisticated data fusion tasks in realtime. Another goal of I-SENSE is to dynamically change the configuration, thus, to be able to react to changes in the systems environment. This paper describes ongoing work in developing necessary hard- and software components in order to perform realtime multi-level data fusion. We present the distributed I-SENSE platform and introduce our multi-level fusion framework. First experimental results on embedded image fusion demonstrates the feasibility of our approach

I-SENSE: A Light-Weight Middleware for Embedded Multi-Sensor Data-Fusion

In our I-SENSE project we demonstrate the combination the scientific research areas multi-sensor data fusion and pervasive embedded computing. The main idea is to provide a generic architecture which supports a distributed data fusion on an embedded system. Due to the high onboard processing and communication power of the used hardware, our proposed architecture is designed to perform sophisticated data fusion tasks. Another goal of I-SENSE research project addresses the reconfiguration of a distributed system at runtime, thus, to be able to react to changes in the systems environment dynamically. This paper though gives an overlook of our developed middleware which eases the development of distributed fusion applications on embedded systems and which includes reconfiguration facilities. We further present some experimental results obtained using our middleware and give an outlook of our ongoing research.

An improved genetic algorithm for task allocation in distributed embedded systems

The main idea of our I-SENSE research project [1] is to provide a generic architecture which supports online data fusion on a distributed embedded system. To accomplish high flexibility, we decided that the functional description of a data fusion system, the so called fusion model, should be defined independently of the present hardware configuration,the so called hardware model. It is the objective, to automatically find a valid mapping of the fusion model to this hardware model. Currently not many publications can be found that report the successful utilization of GP to solve such task allocation problems. Here we present our task allocation method that utilizes GP to find a suitable solution in an adequate time for this optimization problem known to be NP-complete. To improve the performance of the algorithm significantly for our problem domain, we added a few heuristics which are outlined in the following. The fusion model consists basically of a set of communicating tasks which may be represented as a task graph G = (N, E). It is assumed to be a weighted directed acyclic graph, consisting of nodes N = (n1, n2, ..., nm) which represent the fusion tasks and the edges E = (e12, e13, ..., enm) the data flow between those tasks. Each node has some properties, describing the hardware requirements of a task. Every edge indicates the required communication bandwidth between two tasks. The hardware model describes the heterogenous distributed embedded system, consisting of different processors with different properties such as core type, clock frequency and available memory, where the fusion application should run on. Each hardware node has at least one general purpose CPU and optionally some digital signal processors (DSPs) coupled strongly via PCI and I/O ports to connect sensors. Ethernet is used to interconnect the hardware nodes. The traditional genetic algorithm suffers a major drawback if it is applied to our problem: The percentage of valid combinations is usually very small compared to all possible

Task Allocation in Distributed Embedded Systems by Genetic Programming

In this paper we describe a task allocation method, that utilizes genetic programming to find a suitable solution in an adequate time for this NP-complete combinatorial optimization problem. The underlying distributed embedded system is heterogenous, consisting of different processors with different properties such as core type, clock frequency, available memory, and I/O interfaces, interconnected with different communication media. In our applications, which are described as dataflow graphs, the number of tasks to be placed is much larger than the number of processors available. We highlight the difficulties when applying genetic programming to this problem and present our solutions and enhancements, accompanied with some simulation results.