Tech Talk: Scene Motion Detection in Imagery with Anisoplanatic Optical Turbulence
Dr. Richard L. Van Hook, AFRL Sensors Directorate
In long range imaging applications, anisoplanatic atmospheric optical turbulence imparts spatially- and temporally-varying blur and geometric distortions in imagery. The ability to distinguish true scene motion from turbulence warping is important for many image processing and analysis tasks. We present two novel scene-motion detection algorithms specifically designed to operate in the presence of anisoplanatic optical turbulence.
The first method is based on modeling background intensity fluctuations with a Gaussian mixture model (GMM). The GMM parameters are formed using knowledge of the theoretical turbulence tilt variance statistics derived from the Fried parameter or refractive index structure function. Thus, this new method is referred to as the Tilt Variance GMM (TV-GMM) algorithm. While most prior intensity methods use empirical temporal data to estimate a background model, this approach is based on the theoretical atmospheric tilt variance statistics. This technique effectively avoids contamination in the background statistics when true scene motion is present. TV-GMM also considers the application of global image registration as a preprocessing step to improve performance by employing the recently developed residual tilt variance analysis that accounts for image registration.
Rather than forming its statistical model on the (optionally registered) input imagery directly, the second method uses a turbulence simulator. Multiple realizations of atmospheric turbulence are applied to a single prototype background image to create a non-contaminated image stack of the scene background. To incorporate the spatial relationship between neighboring pixels, each pixel's intensity is treated as an independent variable of a single Gaussian distribution. Although full anisoplanatic turbulence simulators are available, alternative approaches are sufficient provided that the anisoplanatic warping is accurate. In this work, a warping simulator based on tilt field statistics is used to provide a fast and reasonable approximation. Because this second method uses a warping simulator, it is referred to as the Warping Simulator Gaussian Model (WS-GM) algorithm.
Both quantitative and qualitative performance analyses are conducted, and the proposed methods are compared against several state-of-the art algorithms using both synthetic and real-world data sets. The three synthetic image datasets are generated with an anisoplanatic numerical wave-propagation simulator that enables per-pixel motion truth. Both TV-GMM and WS-GM outperform the benchmark methods across all turbulence profiles used in this study.
Date and Time
Location
Hosts
Registration
- Date: 31 Oct 2022
- Time: 12:00 PM to 01:00 PM
- All times are (UTC-04:00) Eastern Time (US & Canada)
-
Add Event to Calendar
- Starts 18 October 2022 12:00 AM
- Ends 31 October 2022 12:00 PM
- All times are (UTC-04:00) Eastern Time (US & Canada)
- No Admission Charge
Speakers
Dr. Richard L. Van Hook of U.S. Air Force Research Laboratory
Scene Motion Detection in Imagery with Anisoplanatic Optical Turbulence
In long range imaging applications, anisoplanatic atmospheric optical turbulence imparts spatially- and temporally-varying blur and geometric distortions in imagery. The ability to distinguish true scene motion from turbulence warping is important for many image processing and analysis tasks. We present two novel scene-motion detection algorithms specifically designed to operate in the presence of anisoplanatic optical turbulence.
The first method is based on modeling background intensity fluctuations with a Gaussian mixture model (GMM). The GMM parameters are formed using knowledge of the theoretical turbulence tilt variance statistics derived from the Fried parameter or refractive index structure function. Thus, this new method is referred to as the Tilt Variance GMM (TV-GMM) algorithm. While most prior intensity methods use empirical temporal data to estimate a background model, this approach is based on the theoretical atmospheric tilt variance statistics. This technique effectively avoids contamination in the background statistics when true scene motion is present. TV-GMM also considers the application of global image registration as a preprocessing step to improve performance by employing the recently developed residual tilt variance analysis that accounts for image registration.
Rather than forming its statistical model on the (optionally registered) input imagery directly, the second method uses a turbulence simulator. Multiple realizations of atmospheric turbulence are applied to a single prototype background image to create a non-contaminated image stack of the scene background. To incorporate the spatial relationship between neighboring pixels, each pixel's intensity is treated as an independent variable of a single Gaussian distribution. Although full anisoplanatic turbulence simulators are available, alternative approaches are sufficient provided that the anisoplanatic warping is accurate. In this work, a warping simulator based on tilt field statistics is used to provide a fast and reasonable approximation. Because this second method uses a warping simulator, it is referred to as the Warping Simulator Gaussian Model (WS-GM) algorithm.
Both quantitative and qualitative performance analyses are conducted, and the proposed methods are compared against several state-of-the art algorithms using both synthetic and real-world data sets. The three synthetic image datasets are generated with an anisoplanatic numerical wave-propagation simulator that enables per-pixel motion truth. Both TV-GMM and WS-GM outperform the benchmark methods across all turbulence profiles used in this study.Dr.
Biography:
Dr. Richard L. Van Hook is the Deputy Branch Chief of the Passive RF Sensing branch in the U.S. Air Force Research Laboratory. He received his B.S. and M.S. in Computer Engineering from Wright State University in 2008 and 2014, respectively, and his Ph.D. in Electrical Engineering from the University of Dayton in 2021. His research focuses on atmospheric modeling, turbulence mitigation, and image processing.
Email:
Agenda
12:00-12:05: Introduction
12:05-12:35: Technical Presentation by Dr. Van Hook
12:35-12:45: Q&A
12:45-1:00: Additional time, if needed.