Optimising Construction and Demolition Waste Handling using Computer Vision Techniques

This thesis develops and implements advanced computer vision techniques for efficient segmentation and recognition of construction and demolition waste (CDW), addressing the limitations of manual waste handling at material recovery facilities. A realistic CDW dataset from construction sites was curated, supporting robust evaluation of state-of-the-art deep learning models. To reduce annotation dependence, a semi-supervised adversarial network (DuoSeg++), and a user-interactive system, (PromSeg-Waste), were introduced, enhancing the segmentation performance. Further, the WasteXtract model adapted large-scale vision foundation models for efficient deployment in resource-constrained environments. The release of the CDW-Seg dataset provides comprehensive benchmarking, improving CDW management efficiency, accuracy, and sustainability.