A DNN-Based GLMB Algorithm for Multi-sensor Multi-target Tracking with Dense Group Clutter

Multi-sensor multi-target tracking (MTT) with dense group clutter is facing great challenges in data fusion. One is the high computational complexity caused by the dense group clutter; another is that existing fusion algorithms do not have uniform fusion guidelines. To handle these problems, we present a novel multi-sensor MTT algorithm in this paper, which is based on the generalized labeled multi-Bernoulli (GLMB) filter and distributed deep neural networks (DNN) track fusion under the multi-sensor (MS) system, named the MS-GLMB-DNN algorithm. Firstly, a clutter pre-processing is elaborated, it is based on a CFDP (cluster by finding density peaks) algorithm to reduce the effect of the dense group clutter. Then, the GLMB filter is applied to the DNN track fusion framework for effectively fusing various local tracks. Experimental results indicate that, compared with the commonly used covariance intersection (CI) fusion algorithm, the proposed algorithm can effectively eliminate the effect of clutter on target tracking results, reduce computational complexity, and improve tracking accuracy.