Research Of VSLAM Based On RGB Information Entropy

In order to solve the problem that ORB-SLAM2 is susceptible to the influence of illumination and dynamic objects during the feature matching process,the system performance is degraded,and at the same time,it is prone to scale instability in large scenes.This paper adopts two methods: RGB feature matching algorithm based on information entropy and PROSAC algorithm.The former achieves the purpose of feature matching by generating feature descriptors by comparing the RGB information entropy of the image,and the latter solves the problem of feature mismatching by gradually eliminating mismatched feature points.Combining the two algorithms can improve system accuracy,reduce the chance of feature extraction and matching errors,and improve the clarity of the generated maps.The main working steps are as follows:1)Aiming at the problems that the ORB feature extraction algorithm is prone to unclear texture features and some features are lost during the operation process,a RGB feature matching algorithm based on information entropy is used.Use the RGB feature point search algorithm and RGB feature point descriptor to process any photo with movement,rotation,scale change and illumination change,and then compare it with the photo based on the ORB-SLAM2 algorithm.The experimental results show that the RGB feature extraction algorithm based on information entropy has a good matching effect in the matching of spatial feature points,and the feature recognition is more accurate.In contrast,the ORB-SLAM2 algorithm performs poorly when texture features are unclear or some features are lost.2)In ORB-SLAM2,the algorithm is prone to large matching errors.In order to solve this problem,this paper adopts the feature mismatch elimination algorithm based on PROSAC.Compared with RANSAC used in the ORB-SLAM2 algorithm,the PROSAC algorithm can handle more matching exceptions.Through the elimination work on the data set and the comparison of the real trajectory and the estimated trajectory generated,the results show that during the operation of the entire data set,the predicted trajectory generated by the algorithm used in this paper has a good blue trajectory performance,and the overall algorithm accuracy High,the basic error is between 0.001 and 0.192 m,which means that the error of the algorithm is small,which further verifies the effectiveness of the elimination algorithm in this paper.3)In order to verify the overall performance of the algorithm,the data set was used to compare and verify the algorithm in this paper and the ORB-SLAM2 algorithm.The experimental results show that the algorithm performs significantly better than ORBSLAM2 in terms of the clarity of the generated map when using the fr1＿room and fr1＿xyz sub-datasets.At the same time,the comparison of the relative trajectory error when using the fr1＿room sub-dataset and the fr1＿xyz sub-dataset shows that the error of the feature matching method based on RGB information entropy is much smaller than that based on the ORB feature matching algorithm,and the improvement effect is more than 20% and 100%.More than ten.In addition,the robustness and high precision of the algorithm are further guaranteed by establishing a dense point cloud map.