Robust fusion of GM-PHD filters based on geometric average

•Investigate the problem of the target number underestimation in distributed PHD filters.•Improve the robustness of GA fusion in a missed detection environment.•Design a fusion strategy separately fuse the birth, prediction, and detection components. The Geometric Average (GA) fusion has recently been widely used in multi-sensor multi-target tracking. However, the classical GA-based fusion method performs poorly in the distributed Gaussian Mixture Probability Hypothesis Density (GM-PHD) filters due to the inconsistency of the fused cardinality distribution. Moreover, the performance of the GA fusion method deteriorates further in the presence of missed detections. To this end, this paper presents a Robust Consistency-Balanced GA (RCB-GA) fusion algorithm for the distributed GM-PHD filters. First, we introduce a consistency compensation factor in the fused GM-PHD to dynamically adjust the consistency level of the cardinality distribution according to the spatial proximity of the Gaussian components contained in the local PHDs. Secondly, a three-step fusion strategy separating the fusion of the detection and prediction components is proposed. The robustness of the fusion algorithm is improved by reasonably utilizing the missed detection target information contained in the prediction components of the local PHD. Finally, simulation experiments evaluate the effectiveness and robustness of the proposed algorithm in multi-sensor multi-target tracking scenarios.