Home // ACHI 2026, The Nineteenth International Conference on Advances in Computer-Human Interactions // View article

Vision-Based Estimation of PM2.5 from Surveillance Images

Authors:
Dipti Mitra
Oky Dicky Ardiansyah Prima

Keywords: computer vision; dehazing; semantic segmentation; PM2.5 forecasting; regression.

Abstract:
The paper proposes a vision-based approach for measuring fine Particulate Matter (PM2.5) concentrations by utilizing environmental images as input. A dataset was created by acquiring surveillance camera-captured images from multiple locations in Japan, forming a pair of collected outdoor images and Ground Truth (GT) PM2.5 observation data obtained from monitoring stations. Two preprocessing steps (image dehazing and semantic segmentation) were used to enhance prediction accuracy under varying atmospheric and meteorological conditions. The dehazing method mitigates visual degradation caused by haze, while semantic segmentation extracts and determines object-level information and the coverage amount of extracted objects relevant to PM2.5 estimation. The proposed image-based system combines the dehazed images and segmentation masks, which are then input into a deep learning-based regression model to predict PM2.5 concentrations. The experimental results demonstrate that integrating dehazed images and segmentation masks reduces prediction errors and produces more consistent estimates compared to using original images and other input configurations alone or in combination. The findings indicate that combining enhanced visual representations and segmented objects with deep learning models can serve as an effective and scalable complement to traditional air quality monitoring systems.

Pages: 35 to 40

Copyright: Copyright (c) IARIA, 2026

Publication date: May 24, 2026

Published in: conference

ISSN: 2308-4138

ISBN: 978-1-68558-383-5

Location: Venice, Italy

Dates: from May 24, 2026 to May 28, 2026