Discern Depth Under Foul Weather: Estimate PM2.5 for Depth Inference

Kun Li, Jian Ma, Han Li, Yahong Han, Xibin Yue, Zihao Chen, and Jingyu Yang

Abstract

Nowadays, haze is a common and serious problem and PM2.5 is a main measurement for air quality. Current methods estimate the level of primary pollutant with professional instruments which is expensive and inconvenient. Moreover, with haze, the captured images will be unclear and are difficult to estimate the depth of scene using passive methods. This paper proposes a cheap, fast, and convenient PM2.5 estimation method which only need a captured image using daily-life devices, and further discerns the depth of scene using the estimated PM2.5. We learn haze-relevant classified mapping via hybrid convolutional neural network and combine the high-level features extracted from convolutional layer with ground-truth PM2.5 to train support vector regression (SVR). The transmission map is computed using non-local sparse priors, and the depth map is inferred using the estimated PM2.5 value through the atmospheric scattering model. Experimental results demonstrate that our method achieves accurate PM2.5 estimation and depth inference. This could be very useful in many applications, for both clean and foul weather.

Keywords: Convolutional neural network, PM2.5 estimation, depth estimation, support vector regression (SVR).

Downloads

Code: https://github.com/pmnet-hazy/PMNet

Datasets: https://pan.baidu.com/s/1xuZTWT6gkTm5iTcLyLcwLA password: 7491

Note that the code and datasets can be used only for research purpose. Please cite our journal paper below if you use our datasets or code. Thank you.

Publications