Embedded Vision System for Atmospheric Turbulence Mitigation

Abstract

Outdoor surveillance systems that involve farfield operations often encounter atmospheric turbulence perturbations due to a series of randomized reflections and refraction effecting incoming light rays. The resulting distortions make it hard to discriminate between true moving objects and turbulence induced motion. Current algorithms are not effective in detecting true moving objects in the scene and also rely on computationally complex warping methods. In this paper, we describe a real time embedded solution connected with traditional cameras to both rectify turbulence distortions and reliably detect and track true moving targets. Our comparisons with other methods shows better turbulence rectification with less false and miss detections. FPGADSP based embedded realization of our algorithm achieves nearly 15x speed-up along with lesser memory requirement over a quad core PC implementation. The proposed system is suitable for persistence surveillance systems and optical sight devices.