Scholarlink

<p>This paper introduces a novel 3D multi-object tracker based on tracking-by-detection (TBD) framework. The system leverages parallel feature extraction and data association methods to process 2D appearance features and 3D spatial information, respectively, to achieve accurate tracking of targets in autonomous vehicles and intelligent transportation systems. By combining the Siamese network-based image feature extractor to extract image appearance features and the kinematic model established using Kalman filtering, our tracker effectively utilizes the appearance and spatial information of the object, thus improving tracker accuracy and reliability. Experimental results demonstrate the competitiveness of our tracker on the KITTI tracking benchmark. Compared to previous methods, A parallel feature extraction algorithm is proposed that can independently extract the spatial and appearance information of the object. A novel data association algorithm is designed that makes full use of the spatial information of the object from the point cloud and the appearance information from the image. This work provides substantial technical underpinnings for the advancement of autonomous driving and intelligent transportation technology. <a href="https://github.com/TanShiyu2022/PMTrack">https://github.com/TanShiyu2022/PMTrack</a>.</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

3D multi-object tracking based on parallel multimodal data association

  • 期刊论文

  • 发表日期：

    2025年3月19日

查看全文
  • Shiyu Tan,
  • Xu Li,
  • QiMin Xu,
  • Jianxiao Zhu

摘要

This paper introduces a novel 3D multi-object tracker based on tracking-by-detection (TBD) framework. The system leverages parallel feature extraction and data association methods to process 2D appearance features and 3D spatial information, respectively, to achieve accurate tracking of targets in autonomous vehicles and intelligent transportation systems. By combining the Siamese network-based image feature extractor to extract image appearance features and the kinematic model established using Kalman filtering, our tracker effectively utilizes the appearance and spatial information of the object, thus improving tracker accuracy and reliability. Experimental results demonstrate the competitiveness of our tracker on the KITTI tracking benchmark. Compared to previous methods, A parallel feature extraction algorithm is proposed that can independently extract the spatial and appearance information of the object. A novel data association algorithm is designed that makes full use of the spatial information of the object from the point cloud and the appearance information from the image. This work provides substantial technical underpinnings for the advancement of autonomous driving and intelligent transportation technology. https://github.com/TanShiyu2022/PMTrack.