Research On Light Field Camera Calibration And Active Light Field 3D Reconstruction

The classical imaging system only records the two-dimensional(2D)information of the object,while the quantitative reconstruction of the three-dimensional information of the object requires three-dimensional(3D)measurement technology.Fringe projection profilometry(FPP)is one of the representative techniques for active optical 3D measurement.FPP has many advantages,and the corresponding theories and methods have been relatively complete.However,when performing 3D measurement for specific task,it needs to be further improved in terms of adaptability for complex surface,efficiency of information acquisition and processing,and data integrity.As an emerging technology in the field of imaging,light field imaging can record the direction of light while recording the 2D spatial distribution of light intensity,which breaks through the limitation of classical imaging and provides a new idea for optical 3D measurement.Combined with fringe projection,the active light field 3D measurement technique utilizes phase as the label of light field data,which can effectively realize the light field 3D reconstruction of surface without texture.The main works of this thesis are listed as follows:Combined with fringe projection,phase encoding is introduced into the light field.A digital projector and a light field camera are used to build an active light field 3D measurement system.The phase-encoded light field data is analyzed by using the theory of epipolar plane image(EPI),then the active light field 3D depth estimation can be realized for the surface without texture.However,by analyzing the depth reconstructed results,the reliability and accuracy of light field 3D reconstruction can be significantly affected by the distortion of the main lens in the light field camera.In this thesis,the calibration method for light field camera is designed,which successfully eliminates the distortion of the main lens and achieves high-accuracy 3D reconstruction.The proposed calibration method has relatively simple requirements for devices,which has positive significance for promoting the practical application of light field 3D reconstruction.The main lens distortion of the light field camera is an error source of the light field 3D reconstruction.However,there is a complex coupling between the main lens distortion and the microlens array,which makes it difficult to establish a distortion model explicitly.Arising from the generalized imaging model,we regard the light field camera as a black-box system that provides the mapping between pixels and rays.On this basis,a ray-based model and its calibration method are proposed for light field camera.A flat monitor is used as a calibration target with dense feature points.The pose of the monitor relative to the camera is changed several times,so that the outgoing ray from each pixel of the light field camera has multiple intersection points with the target plane.The imaging process corresponding to the central viewpoint of the light field camera is approximately described with the classical camera model,then the target pose can be estimated.After that,the coordinates of the intersection points are calculated and used for fitting the ray corresponding to each pixel,which provides the ray-based model avoiding from the influence of main lens distortion.Taking arbitrary plane outside the aperture of the light field camera as a reference plane,the intersection points between all the rays and the plane are calculated based on the ray-based model.Then the intersection points are regularly eliminated unevenness to reconstruct the light field distribution that is not affected by the distortion of the main lens.Based on the light field without distortion,the EPI labeled with phase can be generated,and then the slopes corresponding to the homologous points with the same phase value in the EPI are calculated.Combined with the known target poses,the slopedepth map is established to complete the calibration of the light field camera.Based on the above calibration results,the depth information of the object surface can be calculated,and the 3D surface of the object can be further reconstructed combined with the ray-based model.