應用於自動對焦之對焦曲線分析、建模、及轉換

Analysis, Modeling, and Transformation of Focus Profiles for Autofocus

蔡東展(Dong-Chen Tsai)

指導教授 ： 陳宏銘

國立臺灣大學/電機資訊學院/電信工程學研究所/博士(2014年)

https://doi.org/10.6342/NTU.2014.10732

摘要

對於相機及攝影機而言，自動對焦是一個重要的基本功能，它可以讓使用者輕鬆補捉每個瞬間最清晰的影像。設計自動對焦方法有以下三個基本要求：1)準確度、2) 速度、及3)穩定度。通常準確度會影響最後擷取影像的品質；而速度和穩定度會影響最後使用者的感覺和滿意度。目前大部份的自動對焦方法還不夠完善，比如說，以近距離對焦為例，目前大部份相機的自動對焦過程不夠快且不夠穩定，常出現鏡頭來回移動的現象，且最後補捉到影像通常是模糊的。
在本論文中，目標是設計一個快速、準確、及穩定的自動對焦方法。我們認為得到一個有效區間大的對焦曲線模型是設計自動對焦方法一個最根本且重要的問題。一個準確對焦模型可以估計大概合焦位置並使相機做出準確自動對焦決策。根據目前現有對焦曲線模型的觀察，我們把對焦曲線建模之問題重新描述成：決定一個轉換使得所有的對焦曲線轉換後都可以被一個二次式表示。一旦這個轉換決定後，經由反轉換，對焦曲線模型就可以馬上得到。此方法之特色是我們不需要任何對焦評估訊息就可以決定對焦曲線模型。
根據所得到的模型，本論文提出一個有效率靜態自動對焦方法。我們所發展的對焦曲線模型可幫助估算大概合焦位置，做出準確自動對焦決策，且不需經驗性規則及參數。接著，對於動態自動對焦(或連續自動對焦)設計而言，為了避免在開發方法過程中所牽涉到耗時的大量實際測試，本論文提出一個模擬方法來加速演算法開發的過程。我們把動態場景模擬成由許多靜態場景的組合，對於每個靜態場景使用轉換後對焦曲線來表示，並且提出三個常見動態場景之模型且把它們整合到所提出模擬方法中。最後，本論文提出一個連續自動對焦方法。此方法有效避免連續自動對焦常遇到鏡頭前後來回移動的現象並且提升整體自動對焦穩定度。我們使用Kalman濾波器來提升合焦位置估算的準確度並增加整體對焦效能。我們已在各種不同硬體上驗證所開發之自動對焦方法。

關鍵字

自動對焦 ； 相機 ； 攝影機 ； 對焦曲線

並列摘要

Autofocus of a camera is a fundamental function that allows users to capture shape images without human intervention. Three major requirements of autofocus are 1) accuracy, 2) speed, and 3) stability. The accuracy of autofocus affects the quality of the captured image, whereas the speed and the stability of autofocus determine the user experience. The performance of most existing autofocus algorithms is not good enough on these three requirements. For example, when shooting on a close-range object, the autofocus process of most cameras is slow and unstable, and the captured image is usually blurry.
In this dissertation, we aim to design an autofocus algorithm that is fast, accurate, and stable. We identify focus profile modeling as the most fundamental and critical task of autofocus. The proposed focus profile model has large effective range and facilities the estimation of the in-focus lens position and, hence, the autofocus decision. Based on observations of existing focus profile models, we formulate the focus profile modeling problem as that of determining a strictly monotonic transformation that makes a focus profile representable by a quadratic function. Once the transformation is determined, the focus profile model can be readily obtained by the inverse transformation. An appealing feature of the proposed focus profile modeling approach is that the focus profile model can be determined without any knowledge of the focus measurement technique used by a camera.
Based on the derived focus profile model, we propose an efficient autofocus algorithm for still cameras. Our algorithm does not need any heuristics to make autofocus decision and is able to obtain an estimate of the in-focus lens position by only three focus data samples. Next, for the design of continuous autofocus algorithm which often involves repetitive, time-consuming field tests on real scenes, we present a simulation method to expedite the design by replacing laborious field tests with simulation tests. We model scene dynamics by decomposing a dynamic scene into a sequence of static scenes and representing the sharpness responses of the video camera to each static scene by a transformed focus profile. Furthermore, we propose three dynamic models associated with three common dynamic scenes and incorporate these three dynamic models into the proposed simulation framework to obtain effective clues for the refinement of continuous autofocus algorithms.
Finally, to avoid the so-called bouncing phenomenon that is resulted from back-and-forth lens movements around the in-focus lens position and improve the stability of autofocus, we present a novel continuous focus algorithm that effectively improves the accuracy of in-focus lens position estimate for dynamic scenes by a Kalman filter and provides good user experience by a smooth control of the lens movements in the autofocus process. The proposed autofocus algorithm has been validated on various camera platforms.

並列關鍵字

Autofocus ； camera ； video camera ； focus profile

參考文獻

[1] A. Kinba, M. Hamada, H. Ueda, K. Sugitani, and H. Ootsuka, “Auto focus detecting device comprising both phase-difference detecting and contrast detecting methods,” U.S. Patent 5597999, Jan. 8, 1997.

[2] M. Kawarada, “Image-pickup apparatus and control method thereof,” U.S. Patent 024452, Aug. 26, 2010.

[3] R. Yamasaki, “Image sensing apparatus, image sensing system and focus detection method,” U.S. Patent 0045849, Feb. 25, 2010.

[4] Y. Kusaka, “Focus detection device, focusing state detection method and image apparatus,” U.S. Patent 7873267, Jan. 18, 2011.

[5] D. A. Kerr, “Principle of the split image focusing aid and the phase comparison autofocus detector in single lens reflect cameras,” [Online]. Available: http://dougkerr.net/Pumpkin/articles/Split_Prism.pdf