.png){kind=icon}
返回
Tracking multiple construction workers using pose estimation and feature-assisted re-identification model
.png){kind=icon}
PRE
AI
DOI：10.1016/j.autcon.2024.105771.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Tracking construction workers is crucial for ensuring worker safety, productivity, appropriate resource allocation, and regulatory compliance. However, when multiple workers resemble each other or temporary obstructions occur, maintaining accurate identification of individual workers with computer-vision-based tracking techniques is challenging. This paper proposes a multi-worker tracking framework comprising three key components: 1) a pose estimation model that localizes and generates keypoints for each worker, 2) a selective region algorithm with unique visual signatures and a re-identification (ReID) model that extracts features to distinguish workers, and 3) data association techniques that accurately track multiple workers simultaneously. The evaluation results obtained by using the higher-order tracking accuracy (HOTA) and multi-object tracking accuracy (MOTA) metrics on 16 annotated videos demonstrate the effectiveness of the framework. The selective region algorithm, combined with different configurations of trackers and ReID models, achieves an HOTA index of 85.83 % across various scenarios. This pre-emptive intermediation fosters multi-worker monitoring in dynamic construction environments.
Keyword:
Multi-worker tracking
Object reidentification
Pose estimation
Deep learning
期刊
IF:
11.5
论文数:
6.3K
被引数:
4.2W
机构
引用论文
.png){kind=icon}
Deep learning based active learning technique for data annotation and improve the overall performance of classification models基于深度学习的数据标注主动学习技术，提高分类模型的整体性能
PREAI
分享收藏
查看更多内容