Star Sensor: Star Pattern RecognitionAbstract – Star sensors are attractive as attitude sensors for sounding rockets because of their potential to reduce operational costs. Star sensors consist of a CCD camera and an onboard computer. This package is used to perform star image acquisition, star identification, and camera orientation and provides attitude information. This paper documents an investigation in implementing an algorithm to calibrate such a star sensor by analyzing an initial star image and correlating it with an empirical star image generated from a star catalog.I. IntroductionII. ApproachIII. ResultsIV. Conclusions and Future WorkStar Sensor: Star Pattern RecognitionCharly Hermanson and William McMahanDepartment of Electrical and Computer EngineeringClemson UniversityClemson, SC 29634cspanit, [email protected] – Star sensors are attractive as attitude sensors for sounding rockets because of theirpotential to reduce operational costs. Star sensors consist of a CCD camera and an onboardcomputer. This package is used to perform star image acquisition, star identification, and cameraorientation and provides attitude information. This paper documents an investigation inimplementing an algorithm to calibrate such a star sensor by analyzing an initial star image andcorrelating it with an empirical star image generated from a star catalog. I. IntroductionSounding rockets (sub-orbital unmanned launch vehicles) are an important tool for use in highaltitude, upper atmospheric research. Currently, the in-flight guidance system of these launch vehiclesis the Attitude Control System (ACS). The ACS guides the launch vehicle about a pre-programmedpath using a sensor package may that include: magnetometers, 3-axis rate sensors, single axis rategyros, and the Miniature Inertial Digital Attitude System (MIDAS) platform. Unfortunately, thesedevices are high cost and can be unreliable. Often these sensor devices cannot be reused after flight,adding to the operational costs of the sounding rockets. The high-cost of this sensor package has led toa proposal to develop a CCD star sensor guidance system to reduce system costs and increase ACSreliability for night time launch experiments. This paper describes the first step in developing such astar sensor launch vehicle guidance system.Star sensors consist of a CCD camera and an onboard computer used for image processing.Advances in computer technology have made processors that are capable of performing digital imageprocessing algorithms readily available in small packages at a reasonable cost. The star sensorcomponents are particularly advantageous because of their light weight and multi-functionality. Theyallows for a reduced payload volume and mass. Payload size is an important factor with launchvehicles. Additionally, these components should prove to be easier to install and calibrate than theprevious ACS sensor components. The camera in the star sensor is used for star image acquisition. The star image is then sent to theonboard computer which processes it to determine star identification. There are several differentalgorithms for star identification, which try to correlate known star positions from an empirical starcatalog with the unidentified stars in the star image. Some of these algorithms such as the pyramidalgorithm [16] and the triangulation algorithm [7], [15] are automated; while others require a degree ofmanual intervention. Once the stars in an image have been identified orientation and attitude can beinferred based on which stars are in view and how these stars are arranged in the star image.Once the star sensor has been calibrated and the stars successfully identified, it becomes arelatively simple tracking problem to maintain star positional and identity information in subsequentimage frames. However, the initial calibration and first frame star alignment determination is moredifficult as there is little a priori information other than global positioning (GPS) location and courseorientation. This paper describes an investigation of an algorithm that performs this initial calibrationand star identification. Section II describes the design and investigation approach. Section IIIhighlights the results from the work completed. Section IV describes the conclusions that can bedrawn from the work as well as future work to be completed.II. ApproachA. Coordinate systems descriptionRelative familiarity with different coordinate systems is required in order to identify the stars. Two coordinate systems that are most often used in navigation systems are the Local Horizontal coordinate and the Celestial coordinate. The observer’s location is defined in local horizontal coordinates (i.e. longitude and latitude). Latitude is the horizontal division describing the angle between the earth’s equator and a point in the sky. Longitude is the vertical division indicating the directional vector that is parallel to equator (similar to azimuth) referenced at the Greenwich Meridian located in London. The stars locations in the sky are referenced in celestial coordinates (right ascension , declination). Celestial coordinates are similar to local horizontal coordinates with 2 differences: (1) the reference is at the Vernal Equinox (First point of Aries) and (2) the celestial sphere’s axis is offset fromthe earth’s rotational axis. Figure 2-1 illustrates the celestial coordinate system.QuickTime™ and aTIFF (Uncompressed) decompressorare needed to see this picture. QuickTime™ and aTIFF (Uncompressed) decompressorare needed to see this picture.Figure 2-1. Celestial Coordinate System(http://zebu.uoregon.edu/~js/ast221/lectures/lec03.html)B. Generating an Empirical Star ImageThe Smithsonian Astronomical Observatory (SAO) catalog, which comprised 258997 cataloged stars is then filtered to approximately 5200 stars using the appropriate star brightness, time and location via a MATLAB® script. This condensed SAO catalog significantly reduces the computation time and memory required for star identification.From the condensed SAO catalog, an empirical sky image can be generated to match with the actual photograph image. Bright stars identified from the acquired star image can be converted from image coordinate (X, Y) to celestial coordinate. The