Autonomous star field identification for robotic solar system exploration

Abstract

A six-feature all-sky star field identification algorithm has been developed. The minimum identifiable star pattern element consists of an oriented star triplet defined by three stars, their celestial coordinates and visual magnitudes. This algorithm has been integrated with a CCD-based imaging camera. The autonomous intelligent camera identifies in real time any star field without a priori knowledge. Observatory tests on star fields with this intelligent camera are described. I. Attitude determination We describe a new star field identification algorithm that is suitable for implementation on CCD-based imaging cameras. This research has been motivated by the need to increase the autonomy of robotic spacecraft for interplanetary travel to distant planets. With the knowledge of the position of two stars within a field of view of a CCDbased camera pointed in an arbitrary direction, the spacecraft attitude can be determined, using one of the onboard imaging cameras. Two additional instruments, a Sun sensor and a star sensor, spacecraft maneuvers, and time delays due to communications with Earth-based controllers may be eliminated. A star field observed in a camera field of view offers potentially a large number of stars characterized by the measured star coordinates, right ascension (a, ranging from 0 to 360 degrees), declination (6, ranging from 90 to +90 degrees) and brightness (magnitude, in astronomical terminology, ranging from -1.2 to over 20). The minimum identifiable star field consists of an oriented star triplet characterized by the three stars, their celestial coordinates and visual magnitudes. The application of this algorithm in an observatory environment results in a real-time star field identification without a priori knowledge and requires only a small number of reference catalog stars1. The advantages of a small on-board catalog are small memory requirements for storing the catalog data, the small amount of on-board computing power needed, and the increased likelihood that the catalog data is correct. When employing an all-sky catalog, this algorithm is used effectively for whole celestial sphere. 11. Algorithm requirements Star pattern recognition and identification differs from other pattern recognition problems in that stars are imaged by most instrument systems as smeared points. (The field of view is mostly dark.) Thus, the star pattern does not offer the opportunity to generate line features or area features (regions) to be used for object reconsmction as in digital image processing based on segmentation and shading2. Recently, a pattern matching algorithm has been developed for two dimensional lists of position coordinates to be used for non real-time3v4 identification and cataloging of stars recorded on film5. The algorithm for autonomous star pattern identification has been designed for implementation on a high resolution, imaging, CCD-based camera. A camera detects and records an unknown star field distribution imaged in its field of view. The star field is specified by the camera-pointing direction (unknown) and its field of view, as shown in Fig. 1. The number of stars actually detected in the camera field of view depends on the camera exposure (CCD-integration) time. The input to the star field identification algorithm is a list of detected stellar angular coordinates given in the cameracentered, rectangular, coordinate system, and measured stellar magnitudes. The star field identification algorithm employs the truncated and modified Yale bright star catalog617 as a reference table, listing star angular coordinates on the (unit) celestial sphere. For the idealized case when the measured star angular coordinates and their magnitudes exactly correspond to the catalog data, the star field matching problem reduces to only translation and rotation of two patterns with distribution of point sources with different intensities, on a unit sphere. This is illustrated in Fig. 2 for a portion of a celestial sphere projected on a plane. Unlike the other star pattern identification algorithms3v8, the detected celestial coordinates and the measured stellar magnitudes are not assumed to be a subset of the catalog values. I a n . r n a ,--,, INTRODUCED