| 摘要: | 星象儀是一種量測精確姿態的儀器，本論文描述星象儀演算法的基本架構與整合，及提出新的「星辨識」演算法。演算法架構由「影像處理」至「星搜尋與質心計算」，接著進入「星辨識」，最後「計算姿態」。其中「影像處理」與「星搜尋與質心計算」合併，並未探討影像校正的內容。
提出新的辨識方法，是延伸至過去的星辨識方法，參考常見的「一維特徵向量法」 [1]，以及「金字塔法」 [2]。「一維特徵向量法」有簡易搜尋方法，且能輕鬆建立資料庫，但是加入誤差項目與未取得精確相機參數時，穩定性較低。「金字塔法」穩定性高，但對應相機可見星數量越多時，資料庫大小急遽增加。必須簡化資料庫、縮短搜尋時間。改動相機參數時，需花費時間研究資料庫的調整。納入兩種方式的優點，建立出穩定性高的新方法－「分級平均特徵法」。
新方法的特色是針對不同相機拍攝時，影像中最暗視星等的動態範圍變化有高容忍度。表現的結果在誤差項目的少星及位置誤差中有極高的穩定性。因此設定的視星等範圍內，可使用相同的辨識參數及資料庫。最終測試的模擬結果，確認辨識正確率可達99.93％，且高於其他辨識方法。結合追蹤模式時，即使設定嚴苛的測試條件，正確率仍高於99％。
最後電腦模擬測試、實驗室平臺測試、戶外實拍測試三部分，證明星象儀演算法與硬體整合的穩定性。未來可針對模擬失敗或錯誤的視軸，新增額外條件判定，減少或避免辨識失敗或錯誤的發生。另外也需研究動態影像的影響，確認拖曳星點造成的辨識率與精確度的變化。;This paper describes how to organize the algorithm of the star tracker , and proposes a new star identification algorithm. Different cameras have different parameters. Even with the same type of the camera, there are still some differences in the images with the different cameras. If the algorithms do not use the adjusting library and parameters according to the camera parameters, they may be affected or even impossible to complete. The new method closely follows the impact of changes in interest to stars. It is possible to use the same library and parameters to complete work.
The new star identification method, “Group Average Value Method”, extends from two methods, "one-dimensional vector method" [1], and "Pyramid method" [2]. One-dimensional vector method has a simple search method and is easy to build a database, but it is susceptible to the position error, false stars, and lost stars. Its stability is lower than that of the pyramid method. It is stable and fast when it needs to accurately correspond to the camera parameters. When the number of stars in the database of the pyramid method is increases, the size of the database will increase rapidly. It is necessary to simplify the database and increase the speed of the search method. Once the camera parameters are changed, it takes time to study the adjustment of the database. In order to achieve a new method with high stability, the advantages of the two methods are taken out and integrated into a new star identification method, namely the “Group Average Value Method”.
The new method solves the position error. The lost stars or false stars within a certain number, it can work well. It also has a simple method to establish database and search the stars in the database. It can achieve stable work with different camera and dynamic magnitude.
Finally, through the PC simulation test, Laboratory platform test, and real sky test, the stability of the new algorithm was confirmed. Additional conditions can be added in the algorithm which of the simulation failure or error regions. They can reduce or avoid the occurrence of identification failures or errors. In addition, it is necessary to study the impact of dynamic conditions of the images to confirm the changes in recognition rate and attitude accuracy caused by smeared star spots. |
