Non-measurement Camera Calibration Technology Research And System Implementation

Close-range photogrammetry,as an important branch of photogrammetry and remote sensing,is characterized by means of photography to obtain the two-dimensional image of the target object for the calculation of the three-dimensional spatial coordinates.With the rapid development of modern industrial technology,there are many high-cost three-dimensional measurement systems in order to measure the real-time three-dimensional shape of industrial parts accurately and quickly.Based on the principle of close-range photogrammetry,this paper deeply studies the calibration technology of non-measuring camera and improves the key technology of low-cost non-measurement camera for high-precision three-dimensional measurement.The main contents include marking points image processing technology,marking point coding and decoding technology and automatic detection,binocular non-metric camera calibration algorithm.The specific works are as follows:First,as accurate recognition of the marking point in the camera calibration work is a key step in the system to complete the three-dimensional coordinates solution of object,this paper mainly studies the encoding,decoding method and automatic detection of cross calibration target coding and related image processing algorithms.Second,according to different types of calibration plate,this paper utilizes two kinds of camera calibration methods.As the two-dimensional coordinates and the corresponding three-dimensional coordinates can be obtained by data processing for the plane calibration plate,in this paper,the "black box" model of neural network is used to calibrate the camera,and the genetic algorithm is applied to optimizing the thresholds and the weights to improve the stability and training precision of the network.However,processing the images of cross calibration target can only obtain two-dimensional pixels instead of the accurate three-dimensional coordinates of the marking points.Therefore,the paper proposes the binocular stereoscopic multi-view camera calibration method optimized by bundle adjustment method in order to obtain the accurate internal parameters matrix,the distortion parameters and the three-dimensional coordinates.Then,external parameter matrices and relative attitude of left and right cameras are obtained through the rigid transformation between the three-dimensional coordinates of the stereo images.Finally,the accuracy of the reconstructed coordinate values is verified by two camera calibration experiments,which are used to prove the reliability of the algorithms.The appropriate training network or camera's internal and external parameters and distortion parameters can be obtained by the two camera calibration methods proposed in this paper.Camera model parameters and three-dimensional coordinate solution of marking points can be achieved.