PPE & Hazard Detection
YOLO-based detection of personal protective equipment and safety violations on construction sites, with a Streamlit web app and an on-device mobile app.
Overview
Automates construction-site safety checks by flagging missing protective equipment in images, video and live camera feeds.
My role
Computer Vision Engineer
Problem
Safety inspections relied entirely on human guards monitoring multiple CCTV streams, resulting in missed violations and delayed hazard alerts.
Approach
Mobile: React Native + Expo with Vision Camera and react-native-fast-tflite (GPU delegation), overlaying live detections with FPS, latency and RAM stats; models are exported to TFLite, PyTorch and ONNX. Web: a Streamlit app running YOLO11 with pretrained construction-hazard and PPE models from the Hugging Face Hub, supporting uploads, webcam and image URLs with adjustable confidence thresholds.
Results
- 24+ FPS on modern mobile hardware with GPU delegation (mobile app); ~30 FPS on modern hardware (web app).