Research On Feature Points Matching Based On Stereo Vision

Binocular stereovision is simple and reliable since the way in which human's eyes process information is directly simulated. It is valuable in many fields such as the gesture detection and controlling of micro-operation system, robot navigation and aerial survey, three-dimension measurement and virtual reality.In this paper the algorithms of image matching are lucubrated and the computer system of binocular stereovision is established to achieve the stereo matching of feature points of the object.The camera calibration model of the binocular stereovision system is established by simulating through the BP neural network the corresponding relationship between the object in 3D space and its images in 2D plane of the binocular stereovision system, which avoids the errors caused by imperfect mathematic model. The regularization method is used in the training process of BP neural network, which greatly improved the calibration precision and enhanced the popularity of the BP neural network. An auto-adapted algorithm of the edge detection based on Canny operator is applied to extract the feature points so that the image segmentation threshold is auto-adaptive. The detected edge of the goal objects is more integrated and accurate since the character of partial image was took into account. Moreover, the precision and robustness of the algorithm was enhanced since the algorithm of image matching of feature points based on SSDA algorithm, the Epipolar restriction and the parallax gradient restriction is applied. BP neural network is used to calculate the three-dimension coordinates of the feature points of the object after the corresponding relationship of the feature points between the two images is established.The experimental results indicate that the algorithm of image matching of feature points based on SSDA algorithm in this paper could effectively reduce the matching time and improve the accuracy of matching and that the calibration based on BP neural network could improve the accuracy of calibration.