采用环形模板的棋盘格角点检测

曝光过度和镜头畸变将分别导致棋盘格角点分离和角点局部区域不对称现象。现有的角点检测算法通常难以准确提取棋盘格角点。此外,在复杂背景下,现有角点检测算法也极易出现错误。针对上述问题,本文提出了一种图像坐标系下的基于环形模板的棋盘格角点检测算法。首先,通过分析得出棋盘格角点附近的灰度分布应满足一定的对称性和灰度交替性等性质,进而得出环形模板卷积后的图像应满足的性质。而后,利用该性质来定义并提取棋盘格角点,使提取的角点直接达到亚像素精度。实验结果表明：本文提出的棋盘格角点检测算法在曝光过度,镜头畸变和复杂背景情况下,均能取得较好的棋盘格角点检测效果。将该算法应用于实际摄像机标定,实验结果显示重投影误差在0.3个像素以内。

Chessboard corner detection based on circular template

Over exposure and Camera lens distortion often result in chessboard corner separation and chessboard corner local area asymmetry. The existing chessboard corner detection algorithms can not extract corners in such conditions accurately. Besides, the existing chessboard corner detection algorithms will go wrong easily under complex background. To solve these problems, this paper proposes an algorithm based on circular template under image coordinate. We first analyze the gray distribution of the local area of chessboard corner to achieve the properties of convoluted image and define the chessboard corner with those properties. Finally we extract the chessboard corners’ coordinates through those properties and our detected chessboard corners can reach sub-pixel level in merely one step by employing the image coordinate. Experiment results show that our algorithm can achieve better results in both simple backgrounds and complex scenes. Applying our detector to camera calibration, we obtain smaller re-projection error, which is less than 0.3 pixels.