Research On The Landmarks Matching Algorithm Of Dual Camera Digital Industrial Photogrammetry

With the increasingly rapid development of modern ind ustry, digital industrial photography has became the first choice of modern measurements with high precision measurements, non-contact, high working efficiency, portability and so on features. Dual camera digital industrial photogrammetry system is one of the important modes of digital industrial photography system. The feature that it can instantly get the three-dimensional coordinates of the workpiece to be tested determines that it will play an important role in the dynamic measurement and on- line measurement of the representative of modern industrial measurement. And the image matching in dual camera digital industrial photogrammetry is also one of many key technologies in the most important and most difficult steps. The accuracy of image matching directly affects the application of d ual camera digital industrial photogrammetry. In practice, because of the Occlusion, deformation, distortion, noise and other factors in the imaging process, dual camera image matching has been an ill-posed problem more difficult to solve, and currently an universal image matching method has not been found. Therefore, the s tudy of dual-camera image matching has great practical value and theoretical significance. In this paper, the theory and methods of dual camera image matching were studied, the stage of the algorithm was described and experimental results were given.The major work and conclusions of present paper can be summarized as follows:1. Current development, system configuration and key technologies of the dual camera digital industrial photogrammetry system were expounded systematically2. Several problems of dual-camera image matching were expounded and image matching algorithm based on gray, characteristics and round back light reflection mark were analyzed. A dual camera matching algorithm based on invariance relative relationship was put forward in connection with lack of precision, calculation cumbersome inefficient, dependent on the third photograph, timeliness is not strong and many other deficiencies of existing methods. The automatic matching problem of artificial targets was solved by the algorithm in the case of only two images, according to the principle that the relative relationship of adjacent points on smooth transition surface was consistent on two images. This method was achieved by programming and its good matching efficiency and accuracy was also verified through experimental verification.3. The basic process and major issues of how to use RANSAC algorithm to estimate the fundamental matrix and remove mismatching points in image matching were expounded. This paper analyzed the key points of how to improve the matching efficiency and robustness and proposed an improved RANSAC algorithm based on this, in connection with the low e fficiency robustness needs to be improved and many other deficiencies of formerly RANSAC algorithm. This paper chosed a more rational point with the use of previous sampling results at each sampling process with pre- inspection techniques. An adaptive proportional outer point method was used to reduce the sampling frequency and the homography was also used to reduce the iterations and sampling time of this algorithm. This paper abandoned the formerly method that simplely used the match point corresponding epipolar geometry of t he square of the distance to the threshold for comparison to determine the interior point and used a new weighting function to definite interior point legitimately and improved the robustness of the algorithm. The experimental results proved that the algorithm is more efficient than ordinary RANSAC algorithm and the robustness of the algorithm was also improved.