Research On SLAM Technology Based On 3D Vision

The SLAM technology based on 3D vision uses images captured by binocular or RGBD cameras as data input.The SLAM technology is used to create the scene map,and the robot is positioned in real time.At the same time,simultaneous positioning and map creation technologies have developed rapidly and are widely used in unmanned and unmanned vehicles.The research focus of this thesis includes designing a more efficient and robust visual feature extraction and matching algorithm,as well as key techniques such as nonlinear optimization algorithm and map loop detection algorithm.The main content of this article is as follows:1.This thesis focuses on the extraction and matching of image features.A Bayesian grid statistical visual feature module is integrated into the traditional SLAM technology framework.Using classical feature extraction descriptors and Bayesian grid statistics,the matching of feature points and the elimination of false matching points can be accelerated without losing precision.Subsequent experiments show that the SLAM framework of the Bayesian grid statistics algorithm module can achieve higher speed and more robustness than the traditional framework.2.This thesis improves the posture state estimation algorithm for three-dimensional visual SLAM,makes a quantitative analysis of common state estimation models,and points out their respective advantages and disadvantages.In the actual application process,it was found that SLAM has nonlinear interference in the real environment.In this paper,the traditional Kalman filter algorithm is improved and the state estimation and optimization based on the optical speed adjustment method are used.Under certain circumstances,posture state estimation can be performed well.The experimental results show that both the performance and computational complexity are taken into account based on the optical speed adjustment method,and the nonlinear effect of the real scene on the SLAM is excellently achieved.3.This thesis reconstructs the loop detection algorithm.In the SLAM scheme,the front-end adopts the Bayesian grid visual model to obtain a large number of feature points.In order to achieve real-time and engineering optimality,an addition and subtraction method is applied to the traditional loopback detection algorithm.In the loop detection algorithm,a new information entropy rate verification link is added,which can quickly filter and obtain key frames.Finally,the unsupervised learning method is used to construct the keyword frame of key frames.The new loopback detection algorithm is applicable to one of the algorithm schemes in this paper.The rationality of the loopback detection algorithm after reconstruction is verified through experiments,and the new algorithm scheme also maintains the same accuracy as the traditional algorithm model.4.The thesis has completed the creation of 3D maps.A series of tests and comparisons have been made between the proposed algorithm and the classic SLAM technology,which verifies the feasibility and accuracy of the proposed scheme.In the test comparison,this article has been improved in all aspects of the algorithm time,and the robustness has also been improved,the disadvantage is that the memory consumption of this algorithm is the largest,but with the development of the era of memory occupation can be completely used the hardware solves one by one.