Mobile Atmospheric Sensing using Vision Approach

Abstract

Air quality monitoring, especially the atmospheric phenomenon of thick haze, has been an acute problem in most countries and a hot topic in the atmospheric sensing. Recently thick haze occurs more frequently in most cities of China due to the rapid growth of traffic, farming, wildfires, and industrial development. It forms a low-hanging shroud that impairs visibility and becomes a respiratory health threat. Traditionally the dust, smoke, and other particles in relatively dry sky are reported at fixed meteorological stations. The coverage of these sampling stations is limited and cannot accommodate with the emergent incidence of thick haze from industrial pollution. In addition, the visual effect of thick haze is not yet investigated in the current practice. Thick haze appears colorful veil (e.g., yellowish, brownish-grey, etc) in video log images and results in a loss of contrast in the subject due to the light scattering through haze particles. This paper proposes an intuitive and mobile atmospheric sensing using vision approach. Based on the video log images collected by a mobile sensing vehicle, a Haze Veil Index (HVI) is proposed to identify the type and severity level of thick haze from the color and texture perspective. HVI characterizes the overall veil effect of haze spatially. HVI first identifies the haze color from the color deviation histogram of the white-balanced hazy image. The white-balancing is conducted with the most haze-opaque pixels in the dark channel and seed growing strategy. Then pixel-wise haze severity level of atmospheric veil is inferred by approximating the upper veil limit with the dark color of each pixel in a hazy image. The proposed method is tested on a diverse set of actual hazy video log images under varying atmospheric conditions and backgrounds in Wuhan City, China. Experimental results show the proposed HVI is effective for visually atmospheric sensing. The proposed method is promising for haze monitoring and prediction in UAV and satellite remote-sensing images.