|Ph.D Student||Aides Amit|
|Subject||Lightfield Analysis and Recovery of the Atmosphere|
|Department||Department of Electrical and Computers Engineering||Supervisor||PROF. Yoav Schechner|
|Full Thesis text|
The composition of the earth's atmosphere has a vital effect on life on earth. Accordingly, the task of monitoring its contents, i.e. aerosol and cloud droplet concentrations, draws much interest in science and engineering. Much effort is continuously put in developing novel remote sensing systems and algorithms, and in recent years there has been a growing interest in recovering the three-dimensional (3D) structure of the atmosphere, mainly of clouds. Even so, current methods employed in monitoring the atmosphere require expensive equipment, retrieve only partial data (e.g. only precipitation clouds or cloud-top altitudes) or use simplified models (e.g. horizontally uniform atmosphere). To address these shortcomings, this thesis, proposes both novel algorithms and a new remote sensing approach to sensing and analyzing the atmosphere. Our approach comprises a ground-based network of cameras. It is scalable, low cost and enables three dimensional (3D) observations in high spatial and temporal resolution. To leverage these new capabilities, we develop unique reconstruction algorithms. A camera network in conjunction with these algorithms paves the way to 3D recovery of atmospheric scatterer distributions.
Our approach relies on sky imaging from multiple directions and locations. Such setting is similar to tomography. However, common tomography algorithms, as used for medical purposes, assume a controlled setup and simple image formation models that take into account only light extinction without off-axis scattering. In the atmosphere, both assumptions break. The light source, the Sun, is unidirectional and uncontrolled, and any scene of the sky is dominated by scattering. To address this, we use an image formation model based on 3D radiative transfer. We use it to develop tomography algorithms that recover the underlying atmospheric volume accounting both for single-scattering (haze) and multiple-scattering (clouds) situations.
To assess our proposition, we developed, built and deployed a large-scale network of wide-angle cameras. Using this network, we collected a large dataset of sky images taken simultaneously from multiple directions covering a diversity of weather conditions. Our camera network was deployed in conjunction with additional remote sensing equipment: Raman LIDAR, Aeronet sunphotometer and airborne particle spectrometer. These instruments facilitate the calibration of the camera network, provide initialization for the algorithms and give ground truth measurements for validating the reconstruction.
Based on such a comprehensive solution, the sky camera network and reconstruction algorithms enable unprecedented ways to measure nature. Such sensing can complement satellite imagery, be useful to meteorology, make atmospheric scattering tomography realizable, and give new, powerful tools to atmospheric scientists.