Several Key Issues Research On The Infrared Imaging And Depth Estimation Of The Catadioptric Panoramic System

Catadioptric panoramic system is one type of staring wide field of view(FOV)optical system which consists of a curved mirror and a lens.This kind of system can project 360°horizontal FOV and large vertical FOV onto a circular image according to the cylinder projecting principle.Its real-time is better than motion stitching panoramic system and SWAP is better than static stitching panoramic system.As a result,it's the most potential solution for wide FOV imaging currently.In order to broaden the information acquisition channels of catadioptric panoramic system with the further study on it,the related research have been extended from the visible to the infrared band,and from the two-dimensional image to the three-dimensional image acquisition(add scene depth dimension information).This paper will carry out theoretical and experimental research on the issues existed in these extended research such as the compatibility problem between system application requirements and performance requirements,the complex structures,difficult stereo matching,sparse panoramic depth map and so on.The works of the dissertation are listed as follows:(1)A parameter design optimization method based on virtual depth of field for infrared overall well-focused imaging of the catadioptric panoramic system has been proposed.The first step is to design the initial system parameters for meeting application requirements like field of view,system size,single viewpoint and so on.Then,the initial system parameters would be optimized for overall well-focused imaging,and the process is divided into two parts.The first part is to define the virtual depth of field as the depth range of virtual object surface which corresponding to a panoramic scene in a specific vertical FOV.The second part is to include the virtual depth of field into the lens' depth of field.By optimizing some system parameters to meet the requirement,the overall well-focused imaging can be implemented.The characteristic of such a parameter design optimization method is to establish a unified system model for application requirements and overall well-focused imaging requirements.On the basis of this model design result,the actual refractive lens aberration is optimized further by optical design software.As a result,the final catadioptric panoramic system can fulfill the application requirements and overall well-focused imaging requirements simultaneously.Finally,a cooled medium wave infrared catadioptric panoramic system is designed to verify the method above.Moreover,the NETD and MRTD of the system have been tested.(2)A catadioptric panoramic infrared system design model with constant detection range for the full FOV is put forward.Full FOV constant detection range means the two-dimensional scales of the equidistant similar targets on the image plane are the constant.The goal of the constant detection range is to overcome the issues that the useful FOV of the catadioptric panoramic infrared systems is reduced as the system detection range will change with the vertical FOV.The study is extended to the other two innovative works: 1)A general detection range model for the catadioptric panoramic infrared imaging system with center symmetry mirror is put forward;2)A full FOV constant detection range model with distortion compensation has been put forward to reduce the impact of distortion in practical application.(3)A monocular catadioptric panoramic depth estimation method which can obtain the dense panoramic depth map by only one catadioptric panoramic system is presented(Only the visible band is discussed in this paper).This method can solve some issues existed in conventional binocular-based system like complex structure,difficult stereo matching,sparse depth map and so on.The method considers the virtual scene as the intermediate process of panoramic depth estimation.And the virtual scene is the result of panoramic scene nonlinear compression by the curved mirror.The relationship between real panoramic scene and the virtual scene is modeled by caustic which established their distance relation with the curved mirror.The virtual scene depth is obtained from the relative defocus amount between two images which are shot when the camera focus on the front and rear position of the virtual scene.Experiments have verified the theory.However,the panoramic depth map has some problems like noise,non-clear region edge and fuzzy fine structure.(4)In order to solve the issues existed in the panoramic depth map obtained by the monocular catadioptric panoramic depth estimation method,corresponding data processing optimization method is proposed which is guidance by the local information and structure of the original circle projection.The method is divided into three parts according to the guidance ways: 1)A depth map noise removal method called virtual scene segmentation smooth optimization is presented.This method segments the virtual scene into different continuous smooth regions at first,then the region segmentation results are used to guidance the virtual scene depth value in the same region to do the data surface fitting with the aim of removing the error from the outliers and fluctuations of the virtual scene depth value.2)A fine structure preserving method called structure classification regularization method is proposed.This method firstly distinguishes the smooth regions and the hierarchical structure regions of the original circular projection image with its spatial frequency characteristics,then using the spatial distance and the intensity of the neighboring pixels in such two regions individually to guidance the depth relationship of the neighboring pixels of the depth map in the depth estimation iteration process.This method can preserve the high frequency boundary of the fine structure and remove the depth map noise at the low frequency region.3)A weighted joint guidance filtering method based on the scale variability of the circle projection image is proposed for the structure classification regularization method.This method can realize spatial differential texture removal of the original catadioptric panoramic circle projection image and make the same effect of texture removal on the whole circle projection image.Therefore it can prevent the transmission of the texture intensity fluctuations on the original circle projection image onto the panoramic depth map indirectly.