基于激光扫描的鱼眼相机三维标定方法  

Three-Dimensional Calibration of Fisheye Camera Based on Laser Scanner

作　　者：程梦娇 申夏晶 肖江剑[2] 宋文祥[1] 

机构地区：[1]上海大学机电工程与自动化学院,上海200072 [2]中国科学院宁波材料技术与工程研究所,浙江宁波315201

出　　处：《光学学报》2017年第3期253-261,共9页Acta Optica Sinica

基　　金：国家自然科学基金(61273276);国家科技支撑计划(2015BAF14B00);浙江省自然科学基金(R13F020014)

摘　　要：由于鱼眼相机成像存在较大的畸变,采用二维标定板的方法难以在图像边缘区域获得准确可靠的角点,从而导致标定精度下降,而传统的三维标定法存在标定场建造复杂,特征点数目有限等问题。为此,提出一种基于激光扫描的鱼眼相机三维标定方法。该方法首先通过激光扫描仪获取室内标定空间的三维点云图,然后利用尺寸不变特征变换匹配方法得到点云图与待标定相机照片的对应点的匹配关系,并进行分块随机抽样一致性(RANSAC)筛选,再根据对应点的图像坐标和物理坐标进行三维RANSAC筛选,估算最终的鱼眼相机内外参数。与经典的张正友标定法及其改进方法相比,该方法能够获得更多的有效特征点,使标定精度明显改善,从而较好地实现鱼眼相机的畸变校正。该方法简便、精确,有广泛的适用性。Due to the large distortion of fisheye camera images,the two-dimensional calibration board approaches are often difficult to obtain reliable corners around the board boundary,which may lower the calibration accuracy.Traditional three-dimensional calibration methods have some challenges on building a complex calibration field to obtain sufficient number of feature points.Therefore,a three-dimensional fisheye camera calibration method based on the laser scanner is proposed to achieve more accurate results.First,a laser scanner is used to obtain the point clouds of the fixed indoor environment.Second,an improved scale-invariant feature transform matching method is used to detect the corresponding features between point clouds and camera images.A two-round random sample consensus（RANSAC）algorithm is applied to rejecting outliers among the correspondences,and the internal and external parameters of fisheye camera are estimated based on the image and physical coordinates remaining corresponding points after the 3D RANSAC screening.Compared with the classical Zhang Zhengyou calibration method and its improved methods for various cameras,the proposed method can get more effective feature points to make calibration accuracy much higher,which can provide good lens distortion correction for fisheye camera.The proposed method is convenient,precise and extensively applicable.

关 键 词：机器视觉 鱼眼相机标定 激光扫描 尺寸不变特征变换匹配 随机抽样一致性算法 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]