Construction and Demolition Waste Dataset and YOLOv11-L Model for Real-Time Detection and Dynamic Tracking

This repository contains the dataset, trained models, and demonstration video used in the research paper “Dataset Development and Real-Time Detection, Classification, and Dynamic Tracking of Construction and Demolition Waste on a Moving Conveyor Belt.” The dataset (AI waste Sorting_2.v5i.yolov11.zip) includes 1,630 high-resolution images of construction and demolition waste captured under real conveyor motion in a controlled laboratory setup. Each image is annotated into five material classes — Brick, Concrete, Plastic, Reinforcement (steel), and Timber — using the Roboflow platform in YOLO format. The trained YOLOv11-L (YOLOv11l.pt) and YOLOv8-L (YOLOv8l.pt) model weights are provided for benchmarking and replication. The YOLOv11-L model achieved 94.4 % mAP@0.5 and 71.2 % mAP@[0.5:0.95] on the test set, with real-time inference at 15 FPS. The accompanying demonstration video (Example conveyor detection and tracking demonstration.mp4) shows the real-time detection and ByteTrack-based dynamic tracking of multiple C&D waste materials on a moving conveyor. Contents:• AI waste Sorting_2.v5i.yolov11.zip – Dataset (images + labels + YOLO config).• YOLOv11l.pt – Trained YOLOv11-L model weights.• YOLOv8l.pt – Trained YOLOv8-L baseline model.• Example conveyor detection and tracking demonstration.mp4 – Demonstration video.