Abstract:

Considering the limited communicational bandwidth and energy supply in wireless sensor networks (WSNs), a novel framework for quantized track fusion in a robust and distributed way is proposed. Firstly, the local covariance matrices are compressed to find the diagonal lower bound of the matrix. Then the compressed covariance matrix and the target state estimate are quantized using K means, and sent to the fusion center (FC), ie., cluster header (CH). Secondly, a correlation independent robust tracking fusion algorithm, inner ellipsoidal approximation, is proposed in order to attack the unknown or incomplete correlations among local estimates. Finally, an example is included to illustrate the effectiveness of the proposed algorithm. It performs very closely to the existing results while requires much less bandwidth; on the other hand, compared with the random selection of node activation, the target oriented dynamic clustering approach saves energy consumption up to 42%.