Research On Miniature Binocular Endoscope And Binocular Matching Technology

As a mature nondestructive testing instrument,electronic endoscope is widely used in industry and medical treatment.It can break through the limitation of human observation and display clear two-dimensional image information.With the development of computer vision by leaps and bounds in the past 30 years,binocular endoscope,as a new testing instrument,was generated with the combination of endoscopic technology and binocular stereo vision which is the immediate research focus in computer vision.It can simulate human eyes with two cameras to make three-dimensional perception of objects in space.In this paper,a miniature binocular endoscope system is designed and the camera calibration,system composition,feature point matching and stereo matching algorithm of the system are mainly studied.Firstly,this paper introduced the system scheme and hardware module of the endoscope system,and calibrated the internal and external parameters of the two cameras of the system based on pinhole camera model and the calibration method proposed by Zhengyou Zhang.Secondly,in the aspect of feature point matching,this paper proposed a coding method of an orthogonal symmetric local binary mode(OS-LBP)operator and constructed a compact local feature descriptor with good rotation invariance based on the binary operator.On the basis of an improved partition for a feature region based on Cartesian grid,OS-LBP histogram is computed on every subregion while the coding order of OS-LBP operator is determined according to the main direction of the feature point.Then the histograms are concatenated to construct the local descriptor according to the distribution of subregions.It is proved that the descriptor designed in this paper can keep good invariance under rotation and has the advantage of low dimension with guaranteed considerable distinctiveness and robustness.Moreover,in terms of stereo matching algorithm,this paper proposed a local stereo matching algorithm based on area-based matching.Firstly,the cross-based adaptive window is constructed by the color information of pixels and pixel distance between pixels.Then,the composite cost combining sum of absolute differences(SAD)and the Census transform is applied to realize the cost aggregation in the adaptive window,and the initial disparity map is obtained by WTA(Winner-Takes-All)strategy.Furthermore,a disparity optimization method based on region-base vote was proposed to optimize the initial disparity map to get the accuracy dense disparity map.Experimental results show that the stereo matching algorithm proposed in this paper can improve the accuracy of stereo matching as a whole and greatly reduce the error matching rate,especially in disparity discontinuity and high texture areas.Finally,the length and area measurement experiments and 3D reconstruction experiment were carried out with the designed binocular endoscope system.Experimental results show that the average relative error of length measurement is 3.22% and the average relative error of area measurement is 4.69%,which indicates that the endoscope system can effectively carry out 3D measurement at close range.In the 3D reconstruction experiment,the endoscope with the proposed stereo matching algorithm conducted stereo matching for a real scene.The disparity map and point cloud were obtained,and 3D reconstruction of the scene was realized by an assistant software.