基于稀疏字典学习的单目相机标定方法

Monocular Camera Calibration Method Based on Sparse Dictionary Learning

在计算机视觉检测中,广泛应用张正友标定法进行相机标定。但是至少需要三次或更多次地拍摄标定物,这需要烦琐的人工操作,难以实现相机标定的自动化。基于张正友标定法,该文提出了一种基于稀疏字典学习的相机标定方法。该文根据张正友相机标定模型,在相机的不同焦距下,离线获取大量标定图像,计算出不同焦距下的内参参数,并以此构造稀疏字典。当在线使用该相机的时候,根据一幅图像和该稀疏字典进行相机标定,可以获得非常好的内参初始值,再利用极大似然法优化相机参数。通过内参参数对比和重投影法对标定结果进行了评估,实验结果表明,该方法可以得到有效的相机标定内外参数以及畸变参数,实现单幅图像的相机标定,可以提高产品视觉检测的自动化。

In the computer vision detection,Zhang’s calibration method is widely used for camera calibration.However,at least three or more times of shooting calibration objects are needed,which requires tedious manual operation and is difficult to achieve the automa tion of camera calibration.Based on Zhang’s calibration method,in this paper,camera calibration method Based on sparse dictionary learning is proposed.In this paper,according to the camera model of Zhang’s calibration method,many calibration images are acquired offline under different focal lengths of the camera.Then the internal parameter parameters under different focal lengths are calculated,and a sparse dictionary is constructed.In the process of online detection,the camera can be calibrated according to one image and the sparse dictionary,and a good initial value of the internal parameter can be obtained.Then the maximum likelihood method is used to op timize all the camera parameters.The calibration results are evaluated by the comparison of internal parameters and the method of reprojection.The experimental results show that the method can obtain the effective camera calibration internal,external and distortion pa rameters,achieve the camera calibration of a single image,and improve the automation of product vision detection.