Space Freeform Surface Measurement Method Based On Binocular Vision And Photometric Stereovision

The technology of binocular stereo vision measurement has been widely used in industrial production and real life,such as,heteromorphic object measurement,unmanned driving,defect detection,VR and so on.Since the point clouds corresponding to binocular stereo vision only have depth information and do not have normal vector information,it is difficult to describe the characteristics of the object surface and do not have the ability of accurate vision measurement and fine reconstruction of the threedimensional surface.In this thesis,based on the depth information of binocular stereo vision,the surface normal vector information of photometric stereo vision to propose the application method to fuse binocular and photometric stereo in vision measurement and three-dimensional reconstruction is proposed.This thesis mainly faces these problems: calibration of light source direction,recognition and elimination of high-light area and shadow area,estimation of normal vector of scene surface and fusion of binocular stereo vision and photometric stereo vision.Around the above problems,the main work of this thesis is as follows:Firstly,the light source can not form an effective spot in the calibration method based on the mirror target,the laser emitter is used to simulate the parallel light emitted by the light source and then an effective detection spot is formed on the mirror by utilizing the good convergence characteristics of the laser and the better recognition of the RGB image description space.Using the principle of mirror calibration and the centroid coordinates of the spot,the direction information of the light source is finally obtained.Then,an SVM(Support Vector Machine)detection algorithm based on diffuse reflector is designed for the elimination of high-light region.The algorithm realizes the recognition and filtering of high-light region corresponding to diffuse reflector surface.To solve the problem of recognition of shaded areas,firstly,homomorphic filtering algorithm is used to improve the contrast of shaded areas and reduce the overall size of the pixel value of the shaded areas,then the upper limit of the confidence interval of the sample pixel value in the shaded areas is calculated as the threshold.Finally,the recognition and elimination of shaded areas are realized.Secondly,an algorithm for calculating surface normalized vector based on the principle of photometric stereo vision is investigated in this thesis.For the estimation of normal vectors in high-light region,the normal vectors in high-light region are estimated by using the similarity of normal vectors on spherical reflector based in spherical calibration method.In order to estimate the normal vector of the shadow area,we iteratively fill Shadow Region with Mean Value of Non-Shadow Region in the 5*5 templateFinally,in order to fuse the two visual methods,Poisson equation reconstruction is used to fuse the depth information and normal vector information in three-dimensional reconstruction.Soft tool like : OpenCV and PCL are used to program and implement it.In the aspect of vision measurement fusion,this thesis proposes the concept of pixel photosensitive unit corresponding to scene area.We use the method of dividing the X and Y directions of image photosensitive unit and combine the depth information and gradient field information.According to the geometric triangular relationship,multiple depth information is obtained corresponding to the scene target area and then describes the scene area with multiple depth information values.