Reproduce “Construction Site Safety Monitoring and Excavator Activity Analysis System” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sibo Zhang, Liangjun Zhang
Baidu Research developed an integrated vision-based system for real-time construction site monitoring that detects machines and humans, estimates excavator poses, and recognizes actions. The system achieves 95.0% mAP for object detection with YOLOv5x and 93.3% Top-1 accuracy for action recognition, enabling proactive collision warnings and precise productivity analysis.
Key insights
- Utilizing a single-stage detector like YOLOv5 vastly improves object detection speed (up to 34 times faster) and reduces model size while maintaining high accuracy, crucial for real-time deployment in complex construction environments.
- Combining precise excavator pose estimation with working area segmentation enables a robust, interpretable, rule-based approach for real-time action state determination, which is critical for safety-sensitive applications.
- Deep learning-based video models, specifically SlowFast networks, can achieve superior action recognition accuracy for complex excavator activities, outperforming traditional methods and providing objective data for productivity analysis.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.