978-1-4673-1813-6/13/$31.00 ©2013 IEEE 1 Many-Core Computing for Space-based Stereoscopic Imaging Paul McCall Florida Int. University 10555 West Flagler Street, EC 2220 Miami, Florida 33174 305-348-4106 pmcca001@fiu.edu Malek Adjouadi adjouadi@fiu.edu Gildo Torres Clarkson University 8 Clarkson Ave. Potsdam, NY 13699 torresg@clarkson.edu Chen Liu cliu@clarkson.edu Keith LeGrand Missouri University of Science & Technology 400 W 13th St. Rolla, MO 65401 573-576-6811 kal7cd@mst.edu Jacob Darling jed7w2@mst.edu Henry Pernicka pernicka@mst.edu Abstract— The potential benefits of using parallel computing in real-time visual-based satellite proximity operations missions are investigated. Improvements in performance and relative navigation solutions over single thread systems can be achieved through multi- and many-core computing. Stochastic relative orbit determination methods benefit from the higher measurement frequencies, allowing them to more accurately determine the associated statistical properties of the relative orbital elements. More accurate orbit determination can lead to reduced fuel consumption and extended mission capabilities and duration. Inherent to the process of stereoscopic image processing is the difficulty of loading, managing, parsing, and evaluating large amounts of data efficiently, which may result in delays or highly time consuming processes for single (or few) processor systems or platforms. In this research we utilize the Single-Chip Cloud Computer (SCC), a fully programmable 48- core experimental processor, created by Intel Labs as a platform for many-core software research, provided with a high-speed on-chip network for sharing information along with advanced power management technologies and support for message-passing. The results from utilizing the SCC platform for the stereoscopic image processing application are presented in the form of Performance, Power, Energy, and Energy- Delay-Product (EDP) metrics. Also, a comparison between the SCC results and those obtained from executing the same application on a commercial PC are presented, showing the potential benefits of utilizing the SCC in particular, and any many-core platforms in general for real-time processing of visual-based satellite proximity operations missions. TABLE OF CONTENTS 1. INTRODUCTION ................................................ 1 2. INTEL´S SINGLE-CHIP CLOUD COMPUTER ..... 2 3. ALGORITHM OVERVIEW AND IMPLEMENTATION ................................................ 3 4. RESULTS ........................................................... 4 5. CONCLUSIONS AND FUTURE WORK ................ 5 REFERENCES ....................................................... 6 BIOGRAPHY .......................................................... 7 1. INTRODUCTION Research in close proximity operations has been a growing interest of the military and scientific communities. Recently, with the growing need for Space Situational Awareness (SSA), the scope of proximity operations has expanded from cooperative operations to the navigation about both cooperative and uncooperative resident space objects (URSOs). As the name suggests, URSOs are defined as objects which do not communicate with the inspector spacecraft, and range from space debris to adversarial spacecraft. In any case where an inspector spacecraft needs to navigate with respect to a URSO, one or more relative ranging sensors must be used. The ranging sensor under consideration is a visual-spectrum stereoscopic imager consisting of two low-cost, commercial-off-the-shelf (COTS) cameras and an associated processing platform. Space-based vision-based proximity operations possess the inherent advantage of offering an environment in which a single foreground object is in the field-of view of the stereoscopic imager. The cameras are aimed in the same direction, which allows the single foreground object to be isolated without the need for a computationally expensive object detection algorithm. The performance of the stereoscopic imager can be improved with higher resolution cameras. In order to maintain a quick relative navigation solution, the speed of the associated processing platform must get faster as the resolution of the cameras increase.