Research On Fast Corner Extraction Algorithm In Camera Calibration

In the fields of robot navigation, unmanned battle platform and industrial control, the application of vision measurement is expanding and deepening uninterruptedly, due to which, vision system is required to be characterized by on-line, real-time and reliable capability of capturing 3D space information. As the key point of 3D reconstruction, camera calibration is faced with more challenges. The characteristic point extraction becomes the main choke point of vision system to accomplish on-line and real-time camera calibration. The chessboard corner, which is a typical characteristic point for calibration, is taken for example. Aiming at the shortcomings of current chessboard corner extraction algorithm still being unsatisfactory in veracity and real-time property, this thesis did correlative research.On the base of studying usual corner detection algorithm, typical chessboard corner detection and localization algorithm, this thesis proposed automatic corner detection algorithm and corner localization algorithm based on the Hessian matrix with self-adaptive threshold and homography matrix respectively.The automatic corner detection algorithm based on the Hessian matrix with self-adaptive threshold, established a shape operator to distinguish corner between background according to the speciality of Hessian matrix at chessboard corner, designed a method to obtain the self-adaptive threshold of the shape operator, and eliminated the fake corner with the help of chessboard-corner symmetry.The automatic corner localization algorithm based on homography matrix, designed two automatic method to identify and match the chessboard quadrupole points. Homography matrix was calculated via the matched quadrupole points, by the help of which, the implement effect of corner localization was excellent. Precise localization was realized by subpixel corner detecting in the neighbour of initial pixel corner.Experiments were done to validate the proposed automatic chessboard corner extraction algorithm in this thesis. Experiment results shew that the proposed algorithm be of high veracity and real-time property thanks to its simple operation. Took the result of forementioned algorithm to calibrate camera, the fact that the calibration result was similar to traditional one testified the validity of the proposed algorithm.At the end, conclusion was done to summarize the research work of this thesis, and advice of further research was proposed.