基於平面顯示器之自動相機校正方法

Automatic Camera Calibration Using Flat Panel Displays

呂木天(Mu-Tien Lu)

指導教授 ： 莊仁輝 ； 陳永昇

國立交通大學/資訊學院/資訊科學與工程研究所/博士(2019年)

摘要

相機能捕捉現實三維空間中物件的光影並在二維的影像平面呈現，因此在眾多研究領域中(多媒體、機器人、電腦視覺...等)都扮演了關鍵的角色。相機校正在這些領域中是一最為基礎也是最為重要的研究，因其能在二維的平面影像中還原出三維之空間對應關係。現今主流的相機校正流程通常由人工手持一特殊樣式之校正板置於相機前，手動擷取多張不同角度的校正板影像後再進行相機校正，其存在問題有二：
(一) 在人工轉動校正板之階段，由於是人為操作，其校正的結果勢必存在偏差，此偏差在工廠生產階段是不被允許的。又該偏差雖可藉由良好控制各變異因子減小，但操作者必需精通相機校正知識亦或接受過良好校正訓練，對於技術的推廣十分不便。
(二) 目前絕大多數的校正方法均為同時估測相機內外部參數，故相機內部參數之估測會很大程度地相依於相機外部參數之估測，此種校正方法並不符合現實，在物理定義上相機內部參數與外部參數之意義是完全不相同，而在數學上同時對相機內外部參數進行最佳化也會因不同的物理單位而產生偏差，此內外參數之一致性問題將會減低校正結果之強健性。

有鑑於此，本論文提出一系列基於平面螢幕之全自動的相機校正方法解決上述之問題，其優勢包括：
(一) 全自動相機校正系統，不需任何人工干涉，可同時適用於工廠生產階段以及大眾之使用者，不需任何相機校正知識以及預先之教育訓練。
(二) 直接校正相機內部參數，避免內外參數之一致性問題。
實驗結果顯示本論文提出的相機校正方法在強健性以及準確性上皆優於經典的相機校正方法。

關鍵字

相機校正 ； 魚眼相機 ； 雙眼相機

並列摘要

Camera plays an important role in many research fields, e.g., multimedia, robotics, and computer vision, because of its capability of representing a 3D scene with a 2D image. Thus, camera calibration is the most fundamental and important research topic in these fields. Typically, images of a planar are used to represent various relationships between world coordinate system and camera coordinate system in modern camera calibration procedure. There are three steps in a 3-step modern camera calibration procedure:
Step 1: Put a planar pattern in front of the camera.
Step 2: Take a few pictures on the planar pattern at different orientations.
Step 3: Estimate all parameters simultaneously.

The above procedure has two main issues. First, manually adjusting orientation of the planar pattern in Step 2 may cause a bias in calibration results. Second, the associated algebraic manipulation estimates intrinsic parameters and extrinsic ones simultaneously in Step 3, although the former is independent of the latter.

Therefore, in this thesis, a series of novel and automatically camera calibration methods using flat panel displays are proposed to overcome aforementioned issues. Advantages of our methods include:
(I) Our methods are fully automatic without human intervention. These methods are suitable for both manufactories and public. No prerequisites or pre-training are required.
(II) Our methods are capable of directly estimating a specific set of parameters of the camera to avoid consistency issue.
Experimental results show our methods outperform the classic methods in terms of robustness and accuracy.

並列關鍵字

Camera Calibration ； Fisheye Camera ； Stereo Camera

參考文獻

[1] H. Hu and G. Kantor, "Instance selection for efficient and reliable camera calibration," in 2016 IEEE International Conference on Robotics and Automation, 2016, pp. 4335-4340.

[2] S. Ryusuke and Y. Yasushi, "Accurate calibration of intrinsic camera parameters by observing parallel light pairs," in 2008 IEEE International Conference on Robotics and Automation, 2008, pp. 1390-1397.

[3] Z. Zhang, "A flexible new technique for camera calibration," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 22, pp. 1330-1334, 2000.

[4] Z. Zhengyou, "Flexible camera calibration by viewing a plane from unknown orientations," in Proceedings of the Seventh IEEE International Conference on Computer Vision, 1999, pp. 666-673 vol.1.

[5] L. N. Trefethen, Spectral methods in MATLAB vol. 10: Siam, 2000.