Multi-GMTI fusion for Doppler blind zone suppression using PHD fusion

• Published in: Signal processing 2021-06, Vol.183, p.108024, Article 108024
• Main Authors: Wu, Weihua, Sun, Hemin, Huang, Zhiliang, Xiong, Jiajun, Zheng, Mao, Chen, Chen
• Format: Article
• Language: English
• Subjects: Doppler blind zone (DBZ); Ground moving target indication (GMTI); PHD fusion; Probability hypothesis density (PHD); Sensor fusion; Target tracking
• ISSN: 0165-1684; 1872-7557
• DOI: 10.1016/j.sigpro.2021.108024

•we first put forward the idea of using the sensor fusion technique to suppress the DBZ masking problem and verify its feasibility and effectiveness by simulation.•we derive the original PHD fusion based on the generalized covariance intersection (GCI) fusion rule and its efficient Gaussian mixture (GM) implementation.•we fully analyze the cardinality underestimation (CUE) problem of the original PHD fusion through theoretical analysis, examples, and simulations.•we propose an improved PHD fusion algorithm, the core parts of which mainly includes: scaling factor correction for underestimated cardinality, differential treatment of GM components partitioned according to whether they fall into the DBZ or not, and fused label correction to maintain format and display consistency. For ground moving target indication (GMTI) sensor tracking, the existence of the Doppler blind zone (DBZ) seriously deteriorates tracking performance. In order to minimize the adverse effects of the DBZ factor, this paper puts forward the idea of using sensor fusion technique to suppress the DBZ masking problem. First, we derive the probability hypothesis density (PHD) fusion under the generalized covariance intersection (GCI) framework and its Gaussian mixture (GM) implementation for fusing local PHDs from the local trackers. However, we find that there is the problem of cardinality underestimation (CUE) in the original PHD fusion, which is exacerbated when targets are masked by the DBZ. After analyzing this problem in detail, we propose an improved PHD fusion algorithm through operations such as scale coefficient correction, GM component partition, and fused label correction. Finally, the feasibility and effectiveness of the proposed fusion are verified through numerical examples, and it is proved that it alleviates the CUE problem and is significantly better than local trackers.