STIF: A Spatial–Temporal Integrated Framework for End-to-End Micro-UAV Trajectory Tracking and Prediction With 4-D MIMO Radar
STIF: A Spatial–Temporal Integrated Framework for End-to-End Micro-UAV Trajectory Tracking and Prediction With 4-D MIMO Radar
复制标题
DOI：
10.1109/jiot.2023.3244655
复制
发表时间：
2023-11
影响因子：
10.6
通讯作者：
Darong Huang;Zhenyuan Zhang;Xin Fang;Min He;Huizhen Lai;Bo Mi
中科院分区：
文献类型：
--
作者：
Darong Huang;Zhenyuan Zhang;Xin Fang;Min He;Huizhen Lai;Bo Mi
具有随机行为意图的微型无人机的早期轨迹预测有助于消除安全隐患。然而，由于小雷达截面（RCS）的特性，来自微型无人机的后向散射雷达信号可能被淹没在强背景杂波下，导致跟踪失真和错误预测。为此，本文提出了一种基于4维多输入多输出（MIMO）雷达的端到端微型无人机轨迹跟踪与预测的时空集成框架（STIF）。特别是，为了在低信噪比（SNR）条件下获得准确的轨迹，本文将目标检测和跟踪视为相互依赖并共同处理的过程，而不是传统方法中将它们视为两个独立的过程。其优点是，在跟踪的帮助下，可以将原始雷达流中编码的所有连续空间信息合并，以增强连续检测性能，避免仅使用一次扫描就丢失信息。随后，为了适应高机动场景，提出了一个基于意图感知的端到端变压器预测框架，以同时发现隐藏在长期估计轨迹中的空间和时间依赖性。因此，一个四维调频连续波（FMCW）雷达被用来评估所提出的系统。大量的仿真和实验结果表明，STIF优于其他最先进的预测方法，在低信噪比条件下达到了0.3851 m的预测精度。
The early trajectory prediction of micro unmanned aerial vehicles (micro-UAVs) with random behavior intentions facilitates the elimination of potential safety hazards. However, due to the property of a small radar cross Section (RCS), the backscattered radar signals from micro-UAVs may be submerged under strong background clutters, leading to distorted tracking and false prediction. To this end, this article presents a spatial–temporal integrated framework (STIF) for end-to-end micro-UAV trajectory tracking and prediction based on a 4-D multiple-input–multiple-output (MIMO) radar. Especially, to obtain accurate trajectories in low signal-to-noise ratio (SNR) conditions, the target detection and tracking are considered to be interdependent and addressed jointly in this work, rather than treating them as two separate processes in conventional methods. The advantage is that with the assistance of tracking, all consecutive spatial information encoded in raw radar streams can be incorporated to enhance the continuous detection performance, avoiding information loss using only one single scan. Subsequently, to accommodate high maneuvering scenarios, an intention-aware end-to-end transformer-based prediction framework is presented to simultaneously discover both spatial and temporal dependencies hiding in long-term estimated trajectories. Consequently, a 4-D frequency modulated continuous wave (FMCW) radar is utilized to evaluate the proposed system. Numerous simulation and experimental results indicate that STIF outperforms competing state-of-the-art methods and achieve superior prediction performance with the accuracy of 0.3851 m in low SNR conditions.