Illegal roadside waste dumping represents a persistent urban environmental problem, particularly in rapidly growing cities. Conventional monitoring approaches based on manual field surveys and mapping are labor-intensive, costly, and difficult to scale. This study presents an automated framework for detecting and mapping illegal roadside waste using volunteered street-view imagery and zero-shot semantic segmentation with the Segment Anything Model 3 (SAM 3). A dataset of approximately 14,000 Mapillary images from Dhaka, Bangladesh, collected between 2023 and 2025, was used for the analysis. Waste-related objects were identified using SAM 3 with text prompts, eliminating the need for task-specific training data. Detected instances were geolocated and visualized through an interactive web-based application and can be exported in GeoJSON format for further analysis outside of our web-based application. The results demonstrate the potential of combining visual foundation models with street-level imagery for scalable and cost-effective urban waste monitoring in resource-constrained settings. The proposed workflow is transferable to other cities with sufficient street-level imagery coverage and can support evidence-based municipal planning, cleanup operations, and enforcement. The developed code is publicly available at: https://github.com/hcu-cml/roadside-waste-detection-mapping.