Image-based air quality estimation

Abstract

In this paper, we attempt to estimate the outdoor air quality only using images. To address this problem, we mainly collect an available database of high quality outdoor images. We hope this database will encourage further research on image based air quality estimation. Moreover, we perform comprehensive experiments based on this database. We use different hand-crafted features to analyze the appearance variances of outdoor images in different air quality conditions. Results show that the accuracy of meteorological features is much better than that of traditional hand-crafted features. Moreover, in meteorological features, the extinction coefficient indicating the degree of light intensity attenuated by particles performs best with the accuracy of 64.