VISION-BASED TRACKING METHOD OF CONSTRUCTION WORKERS AT NIGHTTIME BY INTEGRATING IMPROVED YOLOv5 AND DEEPSORT
Author(s):
Guofeng Ma, Yiqin Jing, Zihao Huang, Jing Xu, Houzhuang Zhu
This is an open access article distributed under the Creative Commons Attribution License CC BY 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited
Abstract
Nighttime construction, while already frequently occurring, faces safety challenges due to low environmental visibility and personnel fatigue. However, current research on methods for tracking workers mostly focuses on daytime scenes, making it difficult to achieve good results at night. A vision-based method for tracking workers that combines the improved YOLOv5s model with the Deepsort algorithm is proposed in this study. Specially, the illumination image enhancement algorithm is introduced to preprocess the input video sequence to meet the detection requirements in nighttime environments. The performance of the method was evaluated in experiment based on nine testing videos, resulting in the average multiple-object tracking accuracy (MOTA) of 93.15% and multiple-object tracking precision (MOTP) of 97.62%. Simultaneously, the proposed method demonstrated robustness in the face of common multi-object tracking challenges in the experiment.
KEYWORDS:
Construction Worker, Deep Learning, Illumination Enhancement, Nighttime Construction, Vision-Based Tracking