Answers are grounded in this paper and cite their sources.
Purpose This study proposes an on-demand, privacy-compliant Artificial Intelligence of Things (AIoT) framework for real-time detection and mapping of illegal kerbside household bulk waste dumping in urban environments. The system aims to support local councils with timely spatial intelligence for identifying unreported waste accumulation and recurring dumping hotspots. By integrating edge-based deep learning with waste collection vehicles, the proposed approach seeks to improve operational responsiveness, urban cleanliness, environmental sustainability, and evidence-based waste management decision-making while supporting broader smart city and sustainable development objectives. Design/methodology/approach An end-to-end AIoT system was developed using edge computing and deep learning-based object detection. A novel annotated dataset, KerbTrack, was curated to train and evaluate supervised object detection models, including YOLOv4 and DINO, for kerbside bulk waste detection. To assess scalability and generalisation, a large-scale synthetic benchmark, KerbTrack-Syn, comprising 5,114 images generated using a real-image-guided FLUX.1-dev and LoRA pipeline, was also developed. The detection pipeline was deployed on a NVIDIA Jetson AGX Orin edge computer connected to an IMX335 USB camera mounted on waste collection trucks. Detection metadata and geospatial information were transmitted to a web-based dashboard for visualization, hotspot analysis, and verification of reported versus illegal waste dumping incidents. Findings The proposed framework demonstrated the feasibility of real-time kerbside bulk waste detection and mapping using edge AI technologies. Among the evaluated models, YOLOv26 achieved the highest overall performance on the KerbTrack benchmark, attaining an mAP50 of 90%. However, detailed error analysis and large-scale evaluation on KerbTrack-Syn indicated that transformer-based detectors, particularly DINO, exhibited more balanced detection behaviour, superior false-positive suppression, and greater robustness under diverse visual conditions. The developed Minimum Viable Product (MVP) successfully validated practical deployment on waste collection vehicles, demonstrating the operational viability of AIoT-enabled waste monitoring and hotspot mapping. Research limitations/implications The study represents an MVP-scale implementation and evaluation using a custom dataset collected under limited environmental and operational conditions. Model performance may vary across different urban settings, weather conditions, lighting variations, and waste object diversity. Future research should focus on large-scale longitudinal deployments, multi-city validation, model generalisation, and integration with automated reporting systems. The work also highlights broader research opportunities in edge AI, smart waste monitoring, and urban environmental sensing for sustainable smart city applications. Practical implications The proposed system provides councils and waste management agencies with a scalable and privacy-compliant solution for monitoring illegal kerbside dumping in near real time. Automated hotspot identification can improve operational efficiency, optimise resource allocation, and reduce manual inspection requirements. Edge-based processing minimises privacy concerns by transmitting only metadata rather than raw video streams. The framework can also support evidence-based policy development, improve waste collection planning, and enhance responsiveness to community waste management issues within smart city infrastructures. Social implications Illegal bulk waste dumping negatively affects public health, environmental quality, and neighbourhood amenity. The proposed AIoT framework contributes to cleaner and safer urban communities by enabling earlier identification and removal of illegally dumped waste. Improved waste monitoring can reduce environmental pollution, minimise drainage blockage risks, and enhance community wellbeing. The study also supports several United Nations Sustainable Development Goals (SDGs), particularly those related to sustainable cities, responsible consumption, environmental protection, and resilient urban infrastructure through the application of intelligent digital technologies. Originality/value This study presents a privacy-compliant AIoT framework for real-time detection and mapping of illegal kerbside household bulk waste dumping using edge AI integrated with waste collection vehicles. The research introduces the KerbTrack dataset and the large-scale synthetic KerbTrack-Syn benchmark for model development and evaluation. Unlike conventional static monitoring approaches, the proposed mobile and on-demand system combines advanced object detection, edge deployment, geospatial mapping and dashboard-based analytics. The work further provides a comparative evaluation of contemporary convolutional neural network- and transformer-based detectors, offering practical and methodological contributions to smart waste management and AI-enabled urban sustainability research.
No takes yet. Share an insight, caveat, or question.
Iqbal et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: