English  |  正體中文  |  简体中文  |  全文笔数/总笔数 : 62572/95237 (66%)
造访人次 : 2544118      在线人数 : 105
RC Version 7.0 © Powered By DSPACE, MIT. Enhanced by NTU Library & TKU Library IR team.
搜寻范围 查询小技巧:
  • 您可在西文检索词汇前后加上"双引号",以获取较精准的检索结果
  • 若欲以作者姓名搜寻,建议至进阶搜寻限定作者字段,可获得较完整数据
  • 进阶搜寻

    jsp.display-item.identifier=請使用永久網址來引用或連結此文件: https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/122476

    题名: Mobilenet-JDE: A Lightweight Multi-Object Tracking Model for Embedded Systems
    作者: Chi-Yi Tsai;Yu-Kai Su
    关键词: Deep learning;Multi-object tracking;MobileNet;One-shot tracking-by-detection;Joint detection and embedding;Data association
    日期: 2022-02-14
    上传时间: 2022-03-10 12:12:20 (UTC+8)
    出版者: Springer
    摘要: Multi-object tracking (MOT) is one of the most challenging tasks in the field of computer vision. Although many MOT methods have been proposed in the literature, most of them cannot achieve real-time processing performance, especially those running on embedded platforms with limited computing resources. In this paper, we propose a real-time lightweight MOT method based on MobileNet to effectively improve the MOT processing speed. The proposed tracking method consists of a lightweight MOT model and a post-processing module. In the design of the lightweight MOT model, we have enhanced the lightweight object detection model proposed in our previous work by adding an appearance embedding layer. Moreover, we have also proposed a new anchor box design and a novel feature pyramid network (FPN) to improve the tracking accuracy of the proposed method. In the post-processing module, we have proposed a simple filtering method to replace the Kalman filter used in data association processing to accelerate the processing speed. Experimental results show that the proposed MOT method can reach to high processing speeds of 50.5 Frame-Per-Second (FPS) and 12.6 FPS when running on a desktop computer and an embedded platform, respectively. Moreover, the proposed MOT method also provides a competitive tracking performance when compared with the existing MOT methods. These advantages make the proposed method suitable for many applications running on embedded platforms, such as visual surveillance, visual tracking control of mobile robots, human-robot interaction, etc.
    關聯: Multimedia Tools and Applications 81, p.9915–9937
    DOI: 10.1007/s11042-022-12095-9
    显示于类别:[電機工程學系暨研究所] 期刊論文


    档案 描述 大小格式浏览次数



    DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library & TKU Library IR teams. Copyright ©   - 回馈