Scholarlink

<p>When appearance features are unreliable and geometric features are confused by irregular motions, applying conventional Multiple Object Tracking (MOT) methods may lead to unsatisfactory outcomes. To overcome these limitations, we propose a new robust tracker called Enhanced MOT, which can combine the advantages of deep features with a more accurate Kalman state estimation. We introduce two key modules: the Enhanced Intersection over Union (EIoU) and the Adaptive Kalman Filter (AKF). These modules are specifically designed to address association failures caused by irregular motion and the error accumulation of the Kalman filter, while incurring minimal computational overhead. Furthermore, we introduce a simple yet effective IoU-ReID fusion strategy to enhance the tracker’s robustness and employ Gaussian-smoothed interpolation to handle occlusion-induced missing trajectories. Evaluations on the MOT17 test set demonstrate performance improvements over baseline models, achieving a Higher Order Tracking Accuracy (HOTA) of 65.0% and an ID F1 Score (IDF1) of 80.2%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced-MOT: Robust Multi-Pedestrian Tracking with Enhanced IoU and Deep Features Association

  • 期刊论文

  • 发表日期：

    2025年7月1日

查看全文
  • Yuzhe Mu,
  • Lihui Cen,
  • Shichao Kan,
  • Xiaofang Chen

摘要

When appearance features are unreliable and geometric features are confused by irregular motions, applying conventional Multiple Object Tracking (MOT) methods may lead to unsatisfactory outcomes. To overcome these limitations, we propose a new robust tracker called Enhanced MOT, which can combine the advantages of deep features with a more accurate Kalman state estimation. We introduce two key modules: the Enhanced Intersection over Union (EIoU) and the Adaptive Kalman Filter (AKF). These modules are specifically designed to address association failures caused by irregular motion and the error accumulation of the Kalman filter, while incurring minimal computational overhead. Furthermore, we introduce a simple yet effective IoU-ReID fusion strategy to enhance the tracker’s robustness and employ Gaussian-smoothed interpolation to handle occlusion-induced missing trajectories. Evaluations on the MOT17 test set demonstrate performance improvements over baseline models, achieving a Higher Order Tracking Accuracy (HOTA) of 65.0% and an ID F1 Score (IDF1) of 80.2%.