Moving object detection from images distorted by atmospheric turbulence

TLDR: This work proposes a method that combines two independent approaches, non-rigid image registration and background subtraction using Gaussian mixture modeling (GMM), to detect moving objects in turbulent conditions and demonstrates robustness of the proposed approach under varying turbulence conditions.

Abstract: Atmospheric turbulence degrades image with nonuniform geometric deformations and distortions, due to random fluctuations of refractive index over air media. Typical approaches to turbulence removal do not consider moving objects of interest. We propose a method that combines two independent approaches, non-rigid image registration and background subtraction using Gaussian mixture modeling (GMM), to detect moving objects in turbulent conditions. Nonrigid image registration removes geometric distortions and stabilizes overall scene. Then GMM based background subtraction technique is used to detect moving objects. We demonstrate robustness of our proposed approach under varying turbulence conditions using qualitative and quantitative comparisons with existing methods.