基于序贯滤波的三维多模型跟踪算法

A 3D Multi-Model Tracking Algorithm Based on Sequential Filtering

在三维机动目标跟踪领域,单模型跟踪算法已经无法满足高机动目标的跟踪要求,而多模型算法因其能够处理多种不同场景和变化,能够更有效地应对目标机动运动的多样性和不确定性。三维高机动目标跟踪相较于二维机动目标跟踪,因其复杂的空间动态特性,需要更多的模型构成模型集来实现精确跟踪。这种情况导致经典的交互式多模型算法设计非常复杂,计算资源消耗过大。而传统的多模型算法,由于没有考虑模型之间的交互性,其跟踪性能不佳。针对以上问题,借鉴序贯滤波的思想,本文提出了一种新的多模型跟踪架构。这一架构将序贯滤波与多模型算法相结合,将上一时刻多模型的融合结果作为下一时刻每一个模型滤波的输入,这一过程极大地简化了交互式多模型算法中复杂的交互实现过程。同时,针对模型集组合问题,将三维运动解耦为几种二维运动模型和一维运动模型的组合集,形成新的三维空间机动目标跟踪模型集。通过对仿真模拟数据和实测数据的跟踪对比结果表明,相较于传统的多模型算法,本文提出的基于序贯滤波修正的多模型算法在跟踪精度上有显著的提升,且其跟踪精度与交互式多模型算法相当,但在运算时间上则远远优于交互式多模型算法,显著降低了运算资源的需求。

Multi-model algorithms can more effectively deal with the diversity and uncertainty of target maneuvering movements owing to their ability to handle different scenarios and changes,as single-model tracking algorithms can no longer meet the tracking requirements of high maneuvering targets in three-dimensional(3D)maneuvering target tracking.Compared to two-dimensional maneuvering target tracking,three-dimensional high maneuvering target tracking requires more models to form a model set for precise tracking owing to its complex spatial dynamic characteristics.This situation leads to the complex design of classic interactive multi-model algorithms and the excessive consumption of computing resources.However,traditional multi-model algorithms have poor tracking performance because they do not consider the interaction between models.This study proposes a new multi-model tracking architecture based on sequential filtering to address the above issues.This architecture combines sequential filtering with multi-model algorithms,using the fusion result of multiple models from the previous moment as the input for each model filtering in the next moment.This process greatly simplifies the complex interaction implementation process in interactive multi-model algorithms.Concurrently,for the problem of model set combination,the 3D motion is decoupled into a combination of several two-dimensional and one-dimensional motion models,forming a new 3D space maneuvering target-tracking model set.The tracking comparison results of simulation and measured data show that compared to traditional multi-model algorithms,the proposed multi-model algorithm based on sequential filtering correction demonstrates significantly improved tracking accuracy,comparable to interactive multi-model algorithms;however,the proposed algorithm is far better than interactive multi-model algorithms in terms of computation time,significantly reducing the demand for computing resources.