The Self-calibration Algorithm Research On Parallel Binocular Camera

With the rapid development of information technology, the camera calibration as a basic subject, it’s affects is more and more important, such as3D reconstruction of the image, target tracking, satellite navigation and industrial product detection, etc. and all of these tasks can’t be accomplished without the camera calibration. Accurate reference is necessary for the traditional camera calibration, which is an off-line calibration method, but the camera self-calibration method is a online calibration method,which can be applied to various changing scene, so the self-calibration camera method has very important Applicable value and wide application space.The research content of the parallel binocular camera self-calibration algorithm mainly includes on four aspects:the feature point extraction, image feature points matching, obtaining of the fundamental matrix and essential matrix, calculation of internal and external parameters of the camera. The aspect of the feature point extraction:In view of that the cluster phenomenon is easily appeared in traditional Harris corner extraction, This phenomenon is easily emerge wrong matching on subsequent feature point matching, this paper forward a new method,through the strategy of eliminating adjacent point, we can extract more uniform distribution of corners of the images; The aspect of the feature point matching:In view of the traditional matching algorithm accuracy is not high, this paper through the strategy of the Harris of scale invariant feature, on this basis, adopting the bidirectional matching strategy to eliminate some error matching point so as to reduce false matching rate; The aspect of the fundamental matrix extraction:On the base of analyzing the traditional eight-point method to calculate fundamental matrix which has shortcomings such as sensitive to noise, this paper proposes a mean eight-point method to calculate the fundamental matrix, then test and verify fundamental matrix precision through the using of the average residual function; Finally, through calculating equations associated with the fundamental matrix, the camera internal parameters is recovered, then calculating equations associated with the fundamental matrix and the corresponding intrinsic parameters, in solving the essential matrix, and through the essential matrix back out of the camera parameters, essential matrix is calculated, the camera intrinsic matrix is calculated through calculating equations associated with intrinsic matrix.In order to verify the feasibility of the proposed algorithm, this paper builded a software experiment system based on the VC platform, including image preprocessing, Harris corner extraction, Harris point matching, the solution of fundamental matrix and essence matrix and recovering the camera internal and external parameters and so on. The experimental results show that the parallel binocular camera self-calibration algorithm image occasion can satisfy accuracy requirement,which is based on rules and irregular, all of these research laid a good foundation for further researching the binocular cameras vision system.