Risk State Prediction at Construction Sites
VLM-based risk prediction system for detecting unsafe worker behavior and equipment hazards — funded by Doosan Enerbility
Funded by Doosan Enerbility · 2025
Developed a vision-language model (VLM) based risk-state prediction system fine-tuned to detect unsafe worker behavior and equipment hazards from construction site images.
Key Contributions
- Fine-tuned VLM for industrial safety domain
- Detected unsafe worker behavior and equipment hazards from visual inputs
- Built evaluation benchmark for construction site risk classification