Multimedia Sensor Data Fusion Based On Graph Grammar

With the extensively use of multimedia sensors, it established the the MultimediaSensor Networks to describe daily social activities, the ability of real-time processing andeffective utilization of magnanimous multimedia data have became the focus and hotspotin the field of public safety. The considerable multi-sense information afforded by theMultimedia Sensor Networks offers good opportunities for environment monitor andpotential hazard detection. However, how to acquire valuable information frommagnanimous data, to reach comprehensive description of the content understanding andinformation fusion, to detect emergent events and forecast trend, besides to approach topublic hazard warning, are the matters urgently to be solved.Therefore, this thesis combines the graph grammar and data fusion theory, and aimsat the interaction between sensors, then it adopts the quantitative method to define theinteraction relation of sensor nodes, dynamically evolves and analyses the interaction ofsensor network nodes by using directed graph. It aims to obtain the detection sensorydata of the optimal sensors set, based on the established sensor data as well as the fusionframe, combine with judgment and fusion rules, and to achieve the evaluation of thedetected environment. Its specific work mainly includes the following contents:(1) To take spatial dependence and status attribute as reference factors to obtain themaximum relevance optimal sensors set by using the interactions between sensors, thusreducing the abundant redundant detection data which may be produced by the sensornetwork system.(2) To define the evolution rules of graph grammar, and to combine with the sensornetwork interaction relations, finally lead to the directed graph varied with time factor.The graph takes sensor nodes as the diagram nodes and the quantitative interaction ofsensors as the diagram edges, then dynamically demonstrate and analyse the interactionchange between sensor nodes. (3) To establish a platform to process and fuse sensor data with the combination ofthe context information and time nodes, and to complete processing different modal databy introducing the concept of combining the early fusion and late fusion, which is basedon detection and fusion rules, then to adopt hard decision fusio n strategy to deal withhomologous and heterologous sensor data, then finally acquire the results of judgmentevaluation.Through a comprehensive analysis of simulation experiment, it shows that thetheoretical method proposed by the thesis helps to reduce the redundant data ofmultimedia sensor network and optimize the data fusion method which also provides anew way for public environment unsafe factors detection.