Module 3 · Project 2: Helmet Safety Detection (object detection with YOLO)

OBJECTIVE: build an app that finds people with and without safety helmets in images and video, and learn object detection and how to create a labelled detection dataset from scratch. WHAT WE COVER (in project order): • Classification vs detection vs segmentation — what each answers • Collecting images and writing labelling rules • Labelling with bounding boxes in Roboflow / CVAT / Label Studio • Preparing a YOLO dataset: folder structure, label files, data.yaml, splits, quality checks • Detection concepts: bounding box, IoU, confidence, NMS, how YOLO works • Training YOLO with a pretrained model • Evaluation: precision, recall, mAP, confusion matrix, choosing a threshold • Deployment: image, video and webcam app with Gradio + ONNX export AFTER THIS MODULE YOU CAN: • Create and label your own detection dataset end to end • Train, evaluate and improve a YOLO detector • Deploy a detector that works on images and video DATASET: Hard Hat Workers (public, ~7,000 images) + 50–100 images you label yourself. TOOL: Ultralytics YOLO (built on PyTorch).