A Research Of Multi-sensor Distributed Information Fusion Structure And Algorithm Based On Opinion Dynamic

Multi-sensor information fusion is an important subject in current scientific research,among of which the research on fusion structure and algorithm of target tracking fusion system is the most intensively study.In this paper,the fusion structure and the algorithm of target position will be studied and analyzed.First of all,several existing fusion structures(centralized,distributed,hybrid,etc.)are introduced in this paper,and the advantages and disadvantages of the existing fusion structures have been analyzed.With development of information processing capacity of sensor,the processor inside the sensor is divided into two parts: data processor and fusion processor,which are respectively used to generate and fuse the position message of target.Based on network decision mind,the data processor of each sensor sends the acquired target's information to its fusion processor,the fusion processor sends the message to its out-neighbors,at the same time,associates and fuses the information from its in-neighbors,and then sends the fused message to its out-neighbors.Through such a cyclic process,a decentralized multi-sensor fusion structure is thus constructed in the way of the multi-point parallel fusion.Next,the fusion function in fused processor has been designed,by applying the idea of weighted average into the target position fusion.Based on decentralized distributed fusion structure and weighted average fusion function,the distributed discrete-time fusion algorithm is then constructed.It is proven that the fusion algorithm can realize the consensus fusion in a short time if and only if the multi-sensor network has a directed spanning tree.Furthermore,when the sensor network is strongly connected,the consensus result of the fusion algorithm is the weighted average of all sensors' detected information.Later on,introducing random error into the initial detected value,it is assumed that the detected data is a random variable with expected value of the real position of target and the variance of sensors' detection precision.Using the corresponding theory of statistics,the fusion accuracy of algorithm is analyzed theoretically,and the mathematical expression of fusion precision is obtained.And it's found that the fusion accuracy is not lower than the precision of sensor which has lowest detected precision in the network.Combining with Cauchy-Schwarz inequality,the sufficient condition for the highest fusion accuracy is obtained,and it is theoretically proved that the highest accuracy of the fusion algorithm is not lower than the detection accuracy of any sensor.Additionally,how to add the least edges of original sensor network and design the new weight matrix are proposed,so that the new sensor network can achieve the highest fusion accuracy when running the discrete time fusion algorithm.Finally,the convergence experiments under different network are simulated,which verifies the convergence condition of the fusion algorithm.Adding a "blind" sensor into the sensor network,the fusion process is simulated again,and which indicates that the proposed algorithm will reaches consensus even if there is “blind” sensor.An example is given to illustrate how to achieve the highest accuracy,by adding the minimum number of edges and designing the corresponding weight matrix.