MULTI-LAYER TRACK FUSION ALGORITHM BASED ON SUPPORTING DEGREE MATRIX

The random noises of multi-sensor and the environment make observations uncertain and correlative, so the performance of fusion algorithms is reduced by using observations directly. To solve this problem, a multi-layer track fusion algorithm based on supporting degree matrix is proposed. Combined with the track fusion algorithm based on filtering step by step, it uses multi-sensor observations to establish supporting degree matrix and realize multi-layer fusion. Simulation results show its estimation precision is higher than the original algorithm and is increased by 20% around. Therefore, it solves the problem of target tracking further in the distributed track fusion system.

Distributed Radar Target Tracking with Low Communication Cost

In distributed radar,most of existing radar networks operate in the tracking fusion mode which combines radar target tracks for a higher positioning accuracy.However,as the filtering covariance matrix indicating positioning accuracy often occupies many bits,the communication cost from local sensors to the fusion is not always sufficiently low for some wireless communication chan-nels.This paper studies how to compress data for distributed tracking fusion algorithms.Based on the K-singular value decomposition(K-SVD)algorithm,a sparse coding algorithm is presented to sparsely represent the filtering covariance matrix.Then the least square quantization(LSQ)algo-rithm is used to quantize the data according to the statistical characteristics of the sparse coeffi-cients.Quantized results are then coded with an arithmetic coding method which can further com-press data.Numerical results indicate that this tracking data compression algorithm drops the com-munication bandwidth to 4%at the cost of a 16%root mean squared error(RMSE)loss.

DISTRIBUTED KALMAN FILTERING FUSION WITH PACKET LOSS OR INTERMITTENT COMMUNICATIONS FROM LOCAL ESTIMATORS TO FUSION CENTER

This paper considers the distributed Kalman filtering fusion with passive packet loss or initiative intermittent communications from local estimators to fusion center while the process noise does exist. When the local estimates are not lost too much, the authors propose an optimal distributed fusion algorithm which is equivalent to the corresponding centralized Kalman filtering fusion with complete communications even if the process noise does exist. When this condition is not satisfied, based on the above global optimality result and sensor data compression, the authors propose a suboptimal distributed fusion algorithm. Numerical examples show that this suboptimal algorithm still works well and significantly better than the standard distributed Kalman filtering fusion subject to packet loss even if the process noise power is quite large.

A mixed target estimation fusion algorithm based on Gibbs-GLMB and federated filter

Mixed targets are composed of point targets,extended targets,and group targets.The point target can produce one measurement at most,the extended target and the group target can produce multiple measurements,but the sub-goals of the group target have a certain dependency relationship.At this time,the estimated fusion of the group target is converted to the estimated fusion of sub-targets with formation motion structure,and the distance among the sub-targets is very close,which brings difficulties to the estimated fusion of mixed targets.This paper combines the adjacency matrix in graph theory to dynamically model the discernible group target and introduces the concept of deformation.Also,it uses the finite mixture model method to dynamically model the extended target.Then the Gibbs-GLMB algorithm is used to estimate the state and number of the mixed targets.A dynamic detection federated filter fusion algorithm is proposed to fuse the mixed targets state estimates.The effectiveness of the algorithm is verified in the final simulation.