Construction Vehicle Activity Detection in Low-Frequency Surveillance Imagery and Its Relationship to Local Air Quality

Construction vehicles are both central to construction site workflows and major contributors to local air pollution. This paper develops a 3-stage machine learning pipeline that uses sparse on-site surveillance imagery to detect and classify construction vehicle activity and quantify its relationship to ambient air quality. The pipeline comprises a detection model based on the YOLOv9 architecture, a construction vehicle activity classification model (for which two contrasting architectures are tested, including a ViT-based method and an SVM-based model) and a linear regression analysis between detected vehicle activities and local air quality indicators. Despite operating on low-frequency (5-minute interval) imagery under real-world conditions, the proposed models achieve state-of-the-art performance in both detection and activity inference. Regression analysis reveals a statistically significant but limited correlation between vehicle activity and local pollutant concentrations, suggesting the presence of dominant external sources. These findings demonstrate the feasibility of passive, vision-based environmental sensing in constrained urban deployments and open new avenues for integrating ubiquitous computing with sustainable construction monitoring.