使用单目相机的船舶跟踪
计算机视觉与模式识别
2021-08-25 v1 人工智能
机器学习
机器人学
摘要
本文提出了一种仅使用 GPS 数据进行相机标定的新技术。通过利用边界框的位置和大小来估计距离,实现了一种在视频中跟踪平面上移动物体的新方法,在距离相机每 100 米处实现了 5.55 米的平均预测误差。该解决方案可以在边缘端实时运行,在低功耗物联网环境中实现高效推理,同时能够跟踪多艘不同的船舶。
引用
@article{arxiv.2108.10367,
title = {Marine vessel tracking using a monocular camera},
author = {Tobias Jacob and Raffaele Galliera and Muddasar Ali and Sikha Bagui},
journal= {arXiv preprint arXiv:2108.10367},
year = {2021}
}
备注
12 pages, 9 figures, the paper is based on the submission for the AI Tracks at Sea challenge made by the same team taking to a 3rd place in the competition, included in DeLTA 2021 conference proceedings, published on SCITEPRESS Digital Library and available at https://www.scitepress.org/PublicationsDetail.aspx?ID=yzZS+b/VkZ4=&t=1