Real-time stabilization of long range observation system turbulent video

Abstract

Image and video quality in Long Range Observation Systems (LOROS) suffer from atmospheric turbulence that causes small neighbourhoods in image frames to chaotically move in different directions and substantially hampers visual analysis of such image and video sequences. The paper presents a real-time algorithm for perfecting turbulence degraded videos by means of stabilization and resolution enhancement. The latter is achieved by exploiting the turbulent motion. The algorithm involves generation of a "reference" frame and estimation, for each incoming video frame, of a local image displacement map with respect to the reference frame; segmentation of the displacement map into two classes: stationary and moving objects and resolution enhancement of stationary objects, while preserving real motion. Experiments with synthetic and real-life sequences have shown that the enhanced videos, generated in real time, exhibit substantially better resolution and complete stabilization for stationary objects while retaining real motion.

A fast hierarchical patch-based approach for mitigating atmospheric turbulence

“…Dataset A consists of Long Wave IR (LWIR) sequence with frame size of 540×390. The dataset B [3] and dataset C [6] are real-time turbulence affected sequences with frame sizes of 320 × 200 and 504 × 284 respectively. The water tower dataset from [15] has frame size of 320 × 200.…”

“…Atmospheric turbulence is a random phenomenon that distorts an imaging scene non-uniformly [2], [4], [6], [7], [11], [14], [15], and [16]. The major distortions include geometric deformation, space-time varying blur, and diffraction limited blur [15].…”

“…Even though these recent methods ( [15], [16]) are able to remove most of the turbulence they rely heavily on image registration techniques which are computationally quite expensive. Some approaches ( [4], [6]) in the literature mitigate turbulence effects as well as preserve moving objects in videos.…”

“…Thereafter, a fast approach for stabilization of turbulence was introduced by Fishbain et.al. [6]. This approach claims to compensate turbulence motion without affecting the real object motion.…”

“…Then, displacement map is segmented into two classes based on cluster analysis of motion vectors. Though, this preserves moving objects it does not remove local turbulence motion as shown in [6].…”