S.A.F.E. Laboratory

Developed the AI component of a smart laboratory access system that verifies Personal Protective Equipment (PPE) compliance in real time before granting laboratory access through an RFID-enabled attendance workflow.

Laboratory safety inspections are often performed manually, making them time-consuming, inconsistent, and prone to human error before users enter restricted laboratory areas.

Designed and trained a YOLOv5 object detection model for PPE detection, integrated through a Flask API with an IoT-based attendance system using ESP32 and RFID to automate laboratory entry validation.

• Collected and manually annotated a custom PPE dataset to improve detection performance under realistic laboratory conditions.
• Separated AI inference from the web application through a Flask API for modular deployment.
• Integrated the AI model with IoT hardware to enable automated access control instead of standalone object detection.
• Prioritized real-time inference speed while maintaining reliable detection accuracy.

• Collecting sufficient PPE images with diverse lighting conditions and viewing angles.
• Balancing detection accuracy with inference latency for real-time operation.
• Integrating AI inference with RFID-based hardware communication.

• Deploying AI systems requires significantly more engineering than model training alone.
• Dataset quality often contributes more to performance than architectural complexity.
• System integration becomes the primary challenge once the model reaches acceptable accuracy.

• Deploy optimized models using TensorRT or ONNX Runtime.
• Expand PPE categories for broader industrial safety applications.
• Support multiple simultaneous users.
• Implement edge deployment for lower latency.