Research On 3D Reconstruction Technology In Outdoor Scene Based On Binocular Stereo Vision

With the innovation of mobile robot navigation technology,mobile robots in indoor scenes have been unable to meet people's needs well.3D reconstruction technology is the key for mobile robots to realize navigation technology.Its reconstruction accuracy directly determines the recognition and positioning accuracy of mobile robots.Nevertheless,Considering the many troubles caused by the diversity of outdoor scenes and uncertainty,3D reconstruction technology based on binocular stereo vision has been favored by scholars at home and abroad?In outdoor scenes,it has the advantages of wide measurement range,high accuracy,and good real-time performance.Binocular stereo vision calibration technology is the prerequisite for realizing 3D reconstruction technology,and the stereo matching technology in outdoor scenes directly determines the accuracy of 3D reconstruction in outdoor scenes.Aiming at the issue that the direct calibration of the binocular camera using Zhang's calibration method may cause unstable image distortion and serious distortion at the edges,a method for solving the left and right camera parameters and iterative optimization is proposed.Considering the influence of the extraction accuracy,illumination and noise of the feature points on the calibration board on the calibration of the binocular camera,the image preprocessing steps such as Gaussian filtering,median filtering,binarization and smoothing are added to the calibration method.Experiments show that the error of the calibration method described in this article can be up to 0.1718 pixel and the average error is 0.2074 pixel.Aiming at the problems of the complex model of the existing calibration error evaluation method,the indirect evaluation method and the low real-time performance,binocular camera calibration error evaluation method by epipolar constraint is proposed,which uses the degree of matching between the left and right image features to evaluate the calibration error.The accuracy and efficiency of the stereo matching algorithm in outdoor scenes are often limited by the diversification of objects in the scene,the depth often changes,occlusion,too little texture,the existence of noise,etc.This paper studies a stereo matching algorithm based on multi-level optimization.For the sake of guaranteeing the accuracy of stereo matching,Bouguet's stereo rectification is added to the stereo matching algorithm.This paper breaks the idea that similarity evaluation of the traditional stereo matching algorithm often uses only a single cost calculation model,and uses the BT algorithm to perform segmentation processing and multi-directional fusion to calculate the matching cost.Aiming at the real-time issue brought by the cost calculation approach for stereo matching in this paper,a cost calculation optimization approach based on dynamic programming is added.In order to ensure the accuracy of the stereo matching algorithm when it encounters problems such as noise,occlusion,and local optimality,a post-processing step is added to the algorithm in this paper.Finally,in order to verify the accuracy and robustness of the stereo matching algorithm described in this paper when it encounters outdoor lighting changes,12 sets of 3D reconstruction experiments were conducted under 4 different lighting conditions in 3 different outdoor scenes.Experiments show that the method described in this paper still has great accuracy and robustness when outdoor lighting changes.