Loading...
1 results
Search Results
Now showing 1 - 1 of 1
- Detection of Overflowing Waste and Litter near Trash BinsPublication . Paulo, Diogo José dos Santos; Neves, João Carlos Raposo; Proença, Hugo Pedro Martins CarriçoUrban waste management faces growing challenges due to population growth and improper disposal practices, which often lead to overflowing bins and the presence of parasitic waste in public areas. This dissertation proposes an intelligent system for the automatic detection and segmentation of overflowing waste using advanced computer vision techniques. A custom dataset comprising over 7,200 annotated images was collected using fisheye cameras mounted on moving vehicles in real urban environments. Several state-of-the-art segmentation models, such as YOLOv11, YOLACT, and others, were evaluated in terms of precision, recall, and mean average precision (area under the precision-recall curve). To overcome the limitations of traditional RGB-based methods, a novel approach is introduced that combines RGB images with estimated depth and normal surface maps generated from a single RGB image. The interaction of geometric information with the model provides a fusion of information that enhances the model’s ability to distinguish between real waste and background clutter. The proposed system significantly improves segmentation accuracy, achieving up to 47% mAP gains over baseline methods. The results highlight the potential of this approach for real-time urban waste monitoring and contribute a novel dataset and methodology to the field.