Research On Component Technology Applied To The Universal Data Fusion Testbed

An effective system-evaluation approach for multisensor data fusion system is to found a data fusion test and evaluation platform with the modern computer and simulation technology. The universal data fusion testbed has appeared in these years. It provides a building-block-like design mode of testing objects, and facilitates the test and evaluation in the three levels of control parameters, algorithm modules and system structures. This advantage fully satisfies the system-evaluation requirements during the design of data fusion system. Thus, the universal data fusion testbed is considered to be comparably perfect for the data fusion test-evaluation job.Component technology can offer great support to the design work of testing objects on a universal testbed, which can reduce the costs and difficulties of the testbed development. Therefore, the component technology is a software technique propitious to universal testbed. However, the universal testbed is short of comprehensive acquaintanceship at present. The absence of component technology tightly related with the testbed implementation brings on huge problem in application and the lack of implementation foundation for related functions. So the application of component technology in the universal testbed has been obstructed.This dissertation aimed at the abovementioned problems of the component technology application in the universal data fusion testbed. It has analyzed data fusion nodes, testbed function requirements and module decomposition methods that impact the application of component technology, and has constructed fusion component model, domain framework and tested object model suitable for the universal testbed. Furthermore, the parallel scheduling algorithm for testing objects and the component replacement conditions have also been studied. The main contents are as follows:The impacts of factors such as the data fusion nodes and the module decomposition on fusion component have been investigated. A system model for the data fusion nodes has been constructed and the impacts of the multiformity of the node inputs on the reusabilities, processing and interfaces of fusion component have been analyzed. Then it has been pointed out that the multiformity of the node inputs had made the interfaces of fusion component to be combinatorial. The interactive relationships between fusion components have been compared, based on the two methods of data fusion module decomposition. A fusion component model suitable for the universal testbed has been constructed with these analyses. The model fully embodied the combinatorial characteristic of the interfaces and the two interactive relationships between fusion components.The theoretical and practical testing scopes of a data fusion testbed have been compared and detailed analysis was given for the representation and constitution of a domain framework. Taking fundamental elements of component interfaces as basic cells, the mathematical expression of the domain framework which could avoid the redundancy of a priori information has been presented.The approaches and characteristics of data fusion module decompositions were discussed and the two fundamental characteristics of the topological structure of synthetical components were clarified. Then, according these two characteristics, the assembling efficiency of three types of synthetical components were analyzed and the mathematical representation of the aggregated component that had higher assembling efficiency was given, which made up the research foundation for applications of synthetic components in the universal testbed.It was illustrated that the structures of data fusion systems had certain characteristics such as connectivity, feedback, collectivity and distributivity, and their impacts on the topological structure of testing objects were analyzed. In addition, the two elemental types of test and evaluation were compared. The principles and types of combination of evaluation objects in selective evaluations were discussed profoundly. Then the impacts of these combinations on the topological structure of testing objects were analyzed in detail. According to these analyses, the mathematical representation of testing objects was presented and further studies of component replacement conditions had brought forward the mathematical expressions of relative replacement conditions, which could constitute the implemental foundation for algorithm replacement of the testbed.After studies of the parallel characteristic of testing objects, the mathematical representation of testing objects was simplified to a fusion graph. The parallel scheduling problem of testing objects has been described from three aspects including the task model, the multi-processor platform and the scheduling goal. With the feedback of fusion graph in consideration, the requirements and characteristics of topological sorting for parallel prediction were analyzed, and an algorithm with comparatively small computational and storage complexities was proposed for the topological sort of fusion graphs. A parallel scheduling algorithm in accordance with the peculiarity of testing objects was proposed after studies of triggering conditions and distributing manners of tasks in parallel scheduling. Experiments had shown that the scheduling algorithm could make the best use of the computational resources of multi-processor platforms, and effectively shorten the time data fusion tests needed.Aforementioned studies could establish the foundation for extensive applications of the component technique in universal data fusion testbeds. Conclusions of these studies were applied in a simulation test and development platform for multisensor data fusion algorithms and achieved favorable effects.