Research And Application Of The Efficient Calibration Method For Area Array Camera Based On Compressed Sensing

The area array camera because of its advantage of intuitive measurement image has become the most commonly used detection equipment in industrial visual inspection.However,due to the errors in making the camera lens,the image captured by the area array camera cannot accurately describe the mapping of the world coordinates and the pixel coordinates of the spatial point.In addition,due to the aging of camera lens,the errors caused by distortion will become more and more serious.Therefore,camera calibration has become an indispensable processing of industrial visual inspection.The traditional calibration method of Zhang’s has been widely used in the area,shape,size and other industrial visual inspection because of its high flexibility and precision.However,there are many positions of the calibrator to be measured,so the operation efficiency is difficult to support the applications of on-line detection.Therefore,in this paper,we propose an efficient calibration method for area array camera based on compressed sensing.The following studies are carried out:Following aperture imaging principle of area array camera,the world coordinates of spatial points are mapped by perspective projection to pixel coordinates,and a multi parameters non-linear geometric model for camera imaging is established.In view of the difficulty of directly solving the multi parameter nonlinear model,the traditional calibration method of Zhang’s is analyzed in detail.By establishing the underdetermined,positive determined and over-determined solution equations of the initial value of the camera internal parameters,the reason why more than three calibration images in different directions are needed for Zhang’s calibration method to obtain accurate initial value of internal parameters is analyzed.And the importance of initial value of internal parameters for the solution of others parameter is analyzed.Then,an area array camera calibration experiment was conducted based on the Zhang’s calibration method of 3 images and 9 images,and the Zhang’s calibration method showed stable camera parameters only when the calibration image was sufficient.Aiming at the problems that it is necessary to collect a large number calibration images and the stable calibration can’t be carried out with a single calibration image for Zhang’s calibration method,an efficient calibration method based on compressed sensing for area array camera is proposed by combined with the near sparsity of camera parameters’ values displayed on a single basis function.Firstly,Legendre orthogonal polynomial function,Gaussian polynomial function and Chebyshev orthogonal polynomial function are used to represent the camera parameters’ values.Therefore,a sparse representation method of camera parameters based on single basis function is proposed.However,the sparseness of camera parameters’ values based on a single basis function is poor because camera parameters’ values at different focal lengths typically exhibit different distribution characteristics.Thus,a calibration method for area array cameras based on adaptive dictionary basis functions is proposed.This method uses an adaptive dictionary learning algorithm to obtain calibration parameters offline and build a dictionary basis function.In addition,the adaptive dictionary basis function can further updated with the newly obtained camera parameters’ values,which can timely reflect the characterization of camera changes caused by lens aging.Finally,based on the known basis functions and the spatial information provided by one calibration image,in combination with compression sensing theory,the camera parameters are reconstructed and then optimized by iterating.Finally,in this paper,three indicators of the internal parameter solution precision,re-projection error and time consumption were tested and compared with Zhang’s calibration method.The experimental results show that: 1)the camera can be accurately calibrated with only one calibration image,and the time of online calibration can greatly reduce,which is conducive to the online calibration of industrial visual inspection;2)with the increase of the number of calibration checkerboard corner points,more accurate calibration results can be obtained;3)the new calibration result can be used as a new sample to update the dictionary and keep it consistent with the camera state.At the same time,different cameras are used for the same experiment to verify the efficiency and stability of the method.Based on the theoretical algorithm study above,the method proposed in this paper applies to the part size measurement experiments based on binocular visual.Compared with the experimental results of the part size measurement obtained by Zhang’s calibration method,the effectiveness of the proposed method in online detection is verified.