Research On Curved Surface Recognition And 3D Reconstruction Based On Parallel Binocular Stereo Vision

Machine vision technology is widely applied in many fields such as industrial product inspection,precise positioning of medical equipment,road detection and navigation,characteristic object recognition and tracking,with its advantage of non-contact structure,low cost,and convenient implementation.As a result of the continuous development and improvement of image processing technology and computer technology,machine vision technology has realized the function of human eye observation and brain processing information in an increasingly mature attitude,with high detection precision and fast running speed,which fully achieves the requirements of machine automation and intelligence.In machine vision technology,object recognition is a basic research in this field.Effective object recognition plays an important role in the process of understanding targets from images.In this paper,we first studied the recognition method of regular surface,and analyzed the gray distribution of the regular surface's illumination model,which shows the reason why the common method of color recognition or shape detection is not suitable for surface recognition and the advantage of the method of identification.The method is used to extract the gray value of the surface of the object,and the curve fitting error analysis is carried out to determine whether the target object is a curved object according to whether the curve of the fitting curve conforms to the gray curve of the surface.Binocular stereo vision technology is the closest technology to the human eye vision of the three-dimensional perception process,which uses two different observation points and the target scene in the formation of the triangle similarity relation to calculate the spatial coordinates of the target point.The key to this technology lies in the three-dimensional matching.In this paper,we used epipolar geometry principle to improve the calculation method of the initial disparity value based on the confidence propagation matching algorithm.Besides,we calculated the target point matching to the obtained images from both left and right cameras,and acquired the matching point disparity with the help of the regional segmentation algorithm.In the meanwhile,we calculated the three-dimensional coordinates of the target point in the space,and compared it with actual three-dimensional coordinates so as to obtain the error of three-dimensional coordinated recovery based on the basic principle of binocular stereo vision.Finally,we use OpenGL three-dimensional display technology to project three-dimensional coordinates into the space,and add color information to make the target scene reach the visual state.The method of binocular stereo vision is used to acquire the depth information of curved objects,the process of which including camera calibration,image processing,surface recognition,stereo matching,depth recovery and so on.The surface recognition and stereo matching are the main research contents of this subject.In this paper,we established a set of parallel binocular stereo visual experiment platform.The images of different target angles are obtained by using two industrial cameras of the same type.The internal parameters and relative position of the two industrial cameras are calibrated with Zhang Zhengyou checkerboard calibration toolbox,and the calibration accuracy has important impact on the subsequent image processing,stereo matching and deep recovery process.The image processing includes the process of smoothing denoising and edge feature extraction.Then we used the light intensity distribution curve to tell the difference between the plane and surface objects,and adopted the improved confidence propagation stereo matching algorithm to obtain the disparity map of the target scene to finally obtain the threedimensional coordinate information of the target object,thus verifying the feasibility of the method in identifying and measuring the application.