Research On Visual Point Cloud Data Optimization And Map Construction Algorithm

With the continuous advancement of technology,mobile robots are widely used in industrial manufacturing,life services,military and other fields.In the research on intelligent mobile robots,the autonomous positioning and mapping of robots have always been the core issues.SLAM(Simultaneous Localization and Mapping)technology provides a solution to this problem.In SLAM technology,map construction is a very important part.The accuracy and integrity of map construction will directly affect the positioning and navigation accuracy of the robot.At present,among the map construction methods,two-dimensional maps have been studied more and more widely used.However,as robots have higher and higher requirements for scene understanding,three-dimensional maps with more abundant information have also entered the the vision of researchers,construction of efficient and practical 3D visual maps is of great significance for improving the level of robot intelligence.Therefore,this paper will carry out research on the construction of 3D scene visual map based on binocular vision,and carry out research on the key technologies,so as to provide a more accurate and lower cost 3D map construction method for the positioning and navigation of mobile robots.Around this goal,the following three aspects are studied:1.Research on visual feature extraction and matching algorithm.In this paper,the feature extraction algorithm is analyzed,and the ORB feature extraction algorithm is selected to optimize.Aiming at the problem of uneven feature point extraction,a uniform feature point extraction strategy is proposed.Aiming at the problem of mismatching in visual feature matching,a false matching elimination algorithm based on regional motion consistency is proposed,and the effectiveness of the algorithm is verified by experiments.2.Research on visual stereo matching algorithm.Stereo matching is introduced in detail in this paper.Aiming at the problem that the Census algorithm relies heavily on center pixels,which leads to poor robustness of the algorithm,a mean-weighted center pixel replacement method based on neighborhood Gaussian distribution is proposed.Aiming at the problem that the neighborhood information is not fully utilized in the Census algorithm,resulting in low matching accuracy,a neighborhood information cost fusion method is proposed.Improve the accuracy of pixel matching by adding more constraint information.3.Research on point cloud map construction and optimization algorithm.In this paper,the imaging principle of the binocular camera is analyzed,and the depth information of the scene is calculated through the matching relationship,so as to recover the point cloud map of the scene.Then,outlier removal optimization,downsampling optimization and pose optimization are performed on the point cloud map.Aiming at the problem of incremental uncertainty in the L-M algorithm in pose optimization,a method of dynamically adjusting parameters based on reliability is proposed to achieve more stable and fast iteration.In order to verify the effectiveness of the algorithm proposed in this paper,the indoor scene data set is selected to reconstruct the scene in three dimensions,and the performance of the algorithm is analyzed from the qualitative and quantitative dimensions.The experimental results show that the proposed algorithm can effectively reconstruct the scene.,and the accuracy is high.