Integrated Track-to-Track Fusion with Modified Probabilistic Neural Network

Abstract

Utilization of information acquired from a sensor network to improve the tracking accuracy is one of the most important issues in sensor network research. An approach that consists of sensor-based filtering algorithms, local processors and global processor is employed to describe the distributed fusion problem when several sensors execute surveillance over the certain area. For sensor tracking systems, each filtering algorithm utilized in the Reference Cartesian Coordinate System (RCCS) is presented for target tracking when the radar measures range, bearing and elevation angle in the Spherical Coordinate System (SCS). For local processors, each track-to-track fusion algorithm is used to merge two tracks representing the same target. The number of 2-combinations of a set with N distinct sensors is considered for central track fusion. For global processor, a Bayesian method with SemiMarkov process is employed to develop the Modified Probabilistic Neural Network (MPNN) algorithm to compute the weights of local processors. By using local processor weights, the subsequent data fusion is executed to combine the corresponding joint state estimates. Simulation results are presented comparing the performance of the global MPNN algorithm with the local-level fusion algorithm and with the sensor-level filtering algorithm.

Multi-sensor track fusion via Multiple-Model Adaptive Filter

“…True Trajectory with Sensor Locations: The maneuvering target trajectory used in [8], [9] and three fixed sensor platforms are all in the LICCS shown in Fig. 2 The measurement noise correlation matrices for the spherical and Cartesian cases can be related in the following manner:…”

“…The switch based on Bayesian decision theory incorporates with MMAF that can be nearly consistent because it is the self-adjusting variable-bandwidth filter. In particular, the switch is designed by the Modified Probabilistic Neural Network (MPNN) [8] to compute the probability of each IMF to provide the switching capability as adaptive manner to respond the target dynamics. In addition, the switching approach based on MPNN is the advanced version of [9] that the Semi-Markov process is incorporated into Bayesian probabilistic scheme instead of the threshold setting up.…”

Application of Neural Network data associating method in the Radar Network system

“…Due to the inherent parallel structure and self-optimization ability, many kinds of neural networks have been used to solve the data association problem in the multi-target tracking system. Including Boltzmann Machine [7][8] , Multilayer Perceptron (MLP) [9] , Probabilistic Neural Network [10] , Kohonen Neural Network [11] , Hopfield Neural Network [12][13][14][15][16][17] , etc. The computation quantity in learning of Boltzmann Machine's connecting weight increases exponentially with the number of neuron.…”

Distributed Kalman Filtering

“…Perform track fusion optimally for a multiple-sensor system with a specific processing architecture is treated in [295]. Other work cited in Table 2.11 are [338], [15,22,23,24], [16], [17], [19], [21], [20], [71], [72], [73]- [74], [75]- [76], [77], [78]- [79], [124], [126], [260], [261,262], [264,265], [266], [296], [303], [305] and [306]. [266] • Perform track fusion optimally for a multiple-sensor system with a specific processing architecture [295] • Track-to-track fusion for multi-sensor data fusion [296] • Common process noise on the two-sensor fused-track covariance [303] • Track association and track fusion with non-deterministic target dynamics [305] • Comparison of two-sensor tracking methods based on state vector fusion and measurement fusion [306] 2.9.…”