A Gaussian mixture multiple-model belief propagation filter for multisensor-multitarget tracking

• Published in: Signal processing 2024-07, Vol.220, p.109473, Article 109473
• Main Authors: Zheng, Feng, Tian, Yu, Zhan, Weicong, Yu, Jiancheng, Liu, Kaizhou
• Format: Article
• Language: English
• Subjects: Belief propagation; Gaussian mixture; Multiple models; Multisensor-multitarget tracking
• ISSN: 0165-1684; 1872-7557
• DOI: 10.1016/j.sigpro.2024.109473

•The multi-model belief propagation method implemented using Gaussian mixture.•Extracting sensor association measurements using greedy algorithm.•Fusion of all association measurements using the belief iterative updating algorithm.•Multisensor-multitarget three-dimensional tracking simulation scenario with clutter. This paper presents a novel Gaussian mixture multi-model belief propagation (GMM-BP) filter for maneuvering multitarget tracking with multiple sensors. The filter is built upon the BP-based multisensor-multitarget tracking scheme, enabling accurate estimation of target numbers and states. It assumes linear Gaussian target motion, birth process, and sensor measurement models and utilizes the Gaussian mixture model to represent the target's marginal posterior probability density function. Additionally, multiple models are incorporated for target motion, enhancing the maneuvering target tracking capability. Closed-form recursions for means, covariances, and weights of Gaussian components in the filter are derived. To enhance scalability with the number of sensors, a simplified multi-sensor data fusion process is proposed, thereby preserving a linear relationship between Gaussian components and the number of sensors. The performance of the filter was evaluated in a three-dimensional (3D) simulated environment. Compared to the particle-implemented BP filter and the labeled multi-Bernoulli (LMB) filter, the proposed GMM-BP filter exhibited a significant improvement in computational efficiency while maintaining a high level of tracking accuracy.