The Calculation Of The Lunar Elevation Value Of Chang' E3 Based On Binocular Vision

As people explore the moon increasingly,the moon reconstruction technology is more and more important.The accurate reconstruction of the moon helps to further study the lunar topography and environmental resources.Currently,computer vision plays an increasingly important role,of which 3D reconstruction technology based on binocular vision is developing rapidly.3D reconstruction is based on the image acquired by binocular cameras,and then extracts features of the image,and uses the 3D information to reconstruct scene objects.Currently,3D reconstruction technology has been widely used in many fields,such as visual navigation,visual measurement,medical diagnosis,aviation detection and industrial detection.Because of the lack of laser altimeter in Chang'E3,this paper presents an elevation training model based on the lunar HD image of Chang'E3,which combines binocular vision with BP neural network algorithm and uses elevation of Chang'E2 to estimate elevation of Chang'E3.Taking full account of the characteristics of HD moon image of Chang'E3 and its lack of laser altimeter,we combine Chang'E2 and Chang'E3 data for training.The experimental results show that the content of this study is of great practical significance and scientific research value.The main contents of this paper are as follows:1?The method of visual research described in this paper is based on Marr's visual theory.By constructing a parallel binocular system,the object images under a variety of different gestures can be obtained,and then the 3D information can be acquired to realize the reconstruction of the 3D scene according to the relevant stereo matching algorithm and the geometric principle.In this paper,we propose a method to calculate the matching cost of multi-features by fusion color,texture and gradient information.The information of the three feature spaces is used as the matching cost calculation,and the weight parameters are adjusted to get the refined disparity value.The proposed algorithm has high accuracy and low complexity.In this paper,31 pairs of Middlebury website stereo pairs were tested.The results show that the average error of the proposed algorithm is 8.77%,and the average ranking is the best compared with the other nine Middlebury website stereo matching algorithms.This paper also opens a new perspective of stereo matching,which takes into account the information of the multi feature space,thus enriching the internal relations between various images.2?In this paper,we apply binocular vision and machine learning algorithm to lunar data matching and model training of Chang'E2 and 3.The main research contents include stereo matching algorithm of binocular vision,lunar image analysis,lunar image data processing,neural network training model and the calculation of the lunar elevation value.In this paper,we use the SIFT algorithm to extract and match the features of the 2B lunar images of Chang'E2 and 3.Then use PCA algorithm and Hierarchical clustering algorithm to process data of features.And then apply the BP neural network algorithm to train characteristics and elevation values of matching points.Finally,we can quickly get full-month elevation values.The experimental results show that the relative error between the predicted elevation and the actual elevation obtained by the test set is 2.22%,which can quickly and accurately obtain the elevation values of all Chang'E3 images.