针对数据驱动的航迹融合算法精度较低、泛化和适应性差的问题,文中提出了一种数据和模型双驱动的航迹融合算法。该算法主要包含残差提取、残差融合和航迹重建三个部分。残差提取,即模型驱动部分:根据先验知识,建立航迹误差模型,对局部航迹进行近似,并提取其残差作为误差的估计。残差融合,即数据驱动部分:设计了一个多尺度全卷积网络对残差进行融合。航迹重建部分:将网络输出的融合残差还原为融合子航迹,并对连续的多个子航迹进行综合,得到完整的融合航迹。仿真实验表明,该算法不依赖先验信息,融合精度显著优于数据驱动融合算法和传统算法,并具有很好的运动参数和运动模式泛化与适应能力。

To address the problems of low accuracy, poor generalization and adaptability of data-driven track fusion algorithm, a track fusion algorithm driven by data and model is proposed. The algorithm mainly includes three parts: residual extraction, residual fusion and track reconstruction. Residual extraction is the model-driven. The track error model is established based on prior knowledge, and the residual information is extracted using the local track approximation as the error estimation. Residual fusion, that is, the data-driven: a multi-scale fully convolutional network is designed to fuse the track residuals. Track reconstruction: the fusion residual output from the network is restored to the fusion sub-track, and the multiple consecutive sub-tracks are synthesized to obtain the complete fusion track. Simulation results confirmed that the algorithm does not rely on prior information, and the fusion accuracy is significantly better than the data-driven fusion algorithm and the traditional algorithms. Moreover, it has good generalization and adaptability of motion parameters and motion patterns.

TN953+. 7;TP183

国家自然科学基金资助项目(61790552)