Topology-Aware Multi-Object Tracking for Continuous Vehicle Trajectory Extraction Using Distributed Acoustic Sensing
Shixin Wang, Wei Wang
Renmin University of China
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Extracting continuous vehicle trajectories from distributed acoustic sensing (DAS) waterfall plots is challenging because complex driving behaviors and varying vibration intensities can cause signal overlaps. To address this, we propose a two-stage tracking-by-detection framework tailored for highway tunnel environments. First, we analyze several properties of 1D spatiotemporal topology and derive bounds related to state assimilation, trajectory fission, and kinematic decoupling. Second, based on this analysis, we propose a topology-aware Multi-Object Tracking (MOT) mechanism. Instead of using the conventional rigid Hungarian assignment, we introduce a relaxed interval Intersection-over-Union (IoU) matching strategy with historical velocity priors to help decouple intersecting trajectories. Finally, to apply the framework to real-world raw signals, we formulate the real-time extraction of pulse intervals as a 1D sequence segmentation task. After comparing representative neural networks, we select LSTM as the front-end detector. To rigorously evaluate the framework, experiments were conducted on real-world tunnel data and a physics-informed simulation dataset. The simulations involved 20 independent repetitions per scenario, with results reported using 95% confidence intervals to ensure statistical robustness. By comprehensively recording false positives (FP) and false negatives (FN), we demonstrate that our method actively recovers overlapping targets, significantly reducing missed detections. When evaluated on the 1D traffic scenario, the proposed method outperforms adapted 2D state-of-the-art baselines (ByteTrack and OC-SORT) in terms of MOTA and MT metrics, showing improved capacity for maintaining continuous physical trajectories.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
计算机 / AIVideo Surveillance and Tracking Methods
Autonomous Vehicle Technology and Safety · Data Management and Algorithms