| Name: | Avanti Kulkarni |
| Affiliation: | Indian Institute of Technology, Bombay |
| Conference ID : | ASI2024_938 |
| Title : | A Genetic Algorithm based Approach for Tuning Parameters of the Star Tracker Algorithms |
| Authors : | Kudupudi Puja Naga Prasanna, Hrithik Mhatre, Avanti Harshad Kulkarni, Neil George, Hemant Hajare, Chinmay Pimpalkhare |
| Authors Affiliation: | Kudupudi Puja Naga Prasanna, Hrithik Mhatre, Avanti Harshad Kulkarni, Neil George, Hemant Hajare, Chinmay Pimpalkhare (IITB Student Satellite Program, Mumbai - 400076, India) |
| Mode of Presentation: | Poster |
| Abstract Category : | Facilities, Technologies and Data science |
| Abstract : | A star tracker is a highly accurate attitude determination system for satellites. Star tracker’s algorithms rely on certain system-specific parameters to obtain precise quaternions. Tuning these parameters is challenging and time-intensive owing to the wide range of potential parameter values and the need for precise adjustments. This paper proposes the use of a novel approach employing Genetic Algorithm (GA) to identify the optimal parameters for a star tracker’s image processing algorithms. The star tracker algorithm consists of a feature extraction block that determines the centroids of stars in the image, a star matching block that employs a k-vector based technique to match stars with the stars in the guide star catalogue available on-board, and an estimation block that calculates the quaternion using inertial and body frame vectors of the matched stars. The parameters targeted by the GA include: a measure that characterises localization uncertainty which is crucial for effectively exploring the search space and utilising the k-vector search technique in star matching, three validation parameters to authenticate the matched stars, and an error metric for assessing the feasibility of the obtained quaternion. The GA encompasses a fitness function formulated in terms of error between estimated and ideal quaternions in Euler form. The algorithm then uses this fitness function to iterate over the population of parameters, which undergo multiple rounds of crossover, mutation, and selection to reach the optimal values of these parameters. The parameters having comparatively large search intervals were dealt with using logarithmic scaling. Preliminary testing on simulated night sky images has shown promising results, with 80% of the images having errors less than 10 arc seconds. This provides a strong foundation for the algorithm’s performance. Ongoing testing with real night sky images will further validate its practical applicability and robustness in real-world scenarios.
|
