Research On The Monocular Panoramic 3D Sensoring And 3D Reconstruction Based On Mirror Image

Obtaining 3D coordinates on the surface of an object with high efficiency,high accuracy and high robustness is,in 3D measuring and 3D reconstruction,the foundation and key technology.With the development of ultra high definition technology,monocular panoramic sensoring technology based on mirror image,due to its advantages of acquiring rigorous and consistent inner and external camera parameters,applying panoramic vision,and conveniently and simply achieving algorithm of 3D reconstruction during the imaging process,is considered as the technology which possesses broad application prospect.This technology can be applied extensively to detection in agro-industry,robot vision,rapid mould prototyping,virtual realities,medical diagnose and treatment,animation and media,game design,etc.This dissertation focuses on 3D modeling problem of monocular panoramic sensoring based on mirror image,by researching panoramic imaging mechanism about the specular reflection of monocular multi-view and monocular single viewpoint,utilizing optical characteristics of plane mirror set and mirrors with quadratic surfaces,to discover monocular panoramic 3D sensor theory and method which can be applied to different areas.By using monocular panoramic imaging technology,the information on the surface of an object is obtained immediately.By applying perspective unwrapping algorithm,utilizing the constraint of space geometry and of texture color between the views,data from sampling points on the surface of an object are gained.By using 3D reconstruction based on improved B-spline flexible interpolation,improving geometry distortion and operation instability phenomena of traditional reconstruction in the treatment of the non-uniform scan data points,at the same time avoiding the tremendous increase of control points,3D reconstruction of the object tested is realized rapidly.The technology stated above can finally realize monocular panoramic 3D sensoring and 3D reconstruction with low resource depletion and strong real-time processing capacity.This research can provide technological support for automatic,precise and rapid measurement and duplication of the object tested.The main contributions of this dissertation are summarized as follows:? During 3D reconstruction the imaging system with dual-view or multi-view images can not achieve complete modeling by one time.In addition,computational complexity in image matching is high and calibration is complicated,thus a monocular multi-view system and a 3D modeling algorithm based on catadiptric panorama imaging is proposed in this dissertation.Firstly,using stereo visual sensor to obtain a panoramic image of the reconstructed object.By threshold segmentation the image is divided into five basic views.Each view is to be normalized and constraint of space geometry is to be established.Secondly,using discrete voxels to express geometric information of the reconstructed object,according to the optical principle of consistency to judge the surface point of the object.Lastly,using traversing method based on contour voxels to get 3D modeling of the object.This method can get the omni-directional image from different perspectives by one time,using the constraints of space geometry,of texture color and of object curves to quickly reconstruct 3D model of the object,and this method can reduce computational complexity and hardware cost.The practical system has verified the effectiveness of this method.? This dissertation puts forward a 3D reconstructing method of monocular panoramic imaging based rectangular area light path catadioptric.To get more info on the surface of one object and improve the accuracy of reconstruction,we increased virtual viewpoint on the basis of the above study.By using catadioptric transmission mechanism of light path in rectangular area,and applying monocular multi-view sensor,one panoramic image with seven views from different perspectives of geometric object is gained;segmenting this image by threshold value method,extracting foreground information,abtaining outline sample points of the object by using traversal of outline voxels based on polar coordinates,then 3D geometric model of the object is reconstructed.Using monocular multi-view stereo sensor can overcome the problem of usual multi-view images that the inner and outer parameters and color system of different cameras are difficult to keep the same;it can reduce cost of hardware and complexity of multi-view matching calculation,and can increase real-time performance.3D points of the reconstructed object which are obtained by traversal of outline voxels based on polar coordinates are clear and unambiguous.This method has been verified in the actual system.? This dissertation puts forward a 3D reconstructing method of monocular panoramic sensoring based on hyperbolic mirror image.To expand field of view for mirror image and better adapt to the need of mirror imaging system reconstructing the object or scene tested such as pipelines,tunnels,underground facilities,etc,at the same time with the acceptable range of distortion and low hardware cost.This dissertation researches the technology of 3D reconstruction based on monocular panoramic image of hyperbolic mirror on the basis of the above study of 3D reconstruction.By researching on panoramic imaging mechanism about specular reflection of monocular single viewpoint,utilizing optical characteristics of quadratic hyperbolic mirror,one method of 3D measurement and 3D reconstruction based on panoramic sensoring of hyperbolic mirror is implemented.On the basis of monocular panoramic 3D sensor theory to design the panoramic sensoring system with unified camera internal parameters and color system,this system uses imaging technology of hyperbolic mirror to acquire laser reflection points of the object or scene tested,obtaining a panoramic section scanning image.By utilizing the constraint relationship of space geometry,we gain the data from sampling points on the surface of the object tested,using flexible B-spline fitting of sampling points,rapidly achieving 3D reconstruction of the object or scene tested.The technology stated above can be applied to detection and reconstruction of pipeline,tunnel,underground traffic facilities,and so on.It can satisfy panoramic visual real-time measurement and geometric replication in actual engineering,improving the efficiency of panoramic visual geometric entity measurement and reconstruction.? This dissertation puts forward a B-spline flexible interpolating reconstructing method based on point cloud data on monocular panoramic imaging.In the above process of getting point cloud data of the object or scence tested by using the monocular panoramic imaging technology,because sampling points of the cross section are uncertain,the reconstructed 3D model can appear in unexpected shapes,thus the process of compatibility for curves of cross section is needed in the geometric model reconstruction.This process will inevitably increase the amount of data processing,leading to some problems such as low efficiency of operation,geometric shape variation,large increase of control points,etc.Because B-spline has local adjustability and dependencies on node vector,it becomes the main method to fit geometric model of curve and surface.This dissertation takes the five-view imaging as an example.By monocular panoramic vision sensor to get sample points of the geometric object,we calculate parameters based on cross-section sampling points,control flexible spacing range,select candidate knots vector,then get control vertices of 3D object.Lastly,3D geometric shape of the object is reconstructed.This dissertation improves curve and surface shape distortion and operation instability phenomenon of general reconstruction methods in the treatment of the non-uniform cross-section sampling points,effectively reducing the control vertices,attaining smooth and silky entity model to meet the needs of practical engineering.Through the actual system application verification,this method is an effective method for 3D reconstruction.