Research On Several Issues Of Visual SLAM Technology For Mobile Robots

In recent years,with the development of artificial intelligence technology,mobile robots have been widely used in all aspects of social development.Navigation and positioning and environment perception are the basic problems of mobile robots,while simultaneous positioning and map construction(Simultaneous Localization and Mapping,SLAM)are key technologies to realize robot navigation.When the sensor used is a camera,it is called visual SLAM.Compared with SLAM based on lidar,it is simpler and lighter,and the acquired environmental information is richer.Research hotspots in recent years.This article has carried out in-depth research around several issues in visual SLAM technology,and its main research work is as follows:(1)Aiming at the problem of low positioning accuracy of traditional visual odometer method in weak texture environment,this research proposes a visual odometer calculation method based on point and line features.First,improve the ORB algorithm so that the feature points can be evenly distributed in the image,which specifically includes dividing an adaptive size grid on the image,extracting ORB features from the grid,using a quad-tree structure to store the key points,and passing restrictions The depth of the quadtree improves the efficiency of the algorithm;then the LSD line segment features in the environment are extracted;finally,the camera pose is estimated according to the reprojection error model of the point and line features.The experimental results show that the method in this paper can still achieve a good positioning effect in a weak texture environment.(2)Aiming at the problem of false positives in traditional loop detection methods,a loop detection method based on multi-region feature weighting and twin network is proposed.First,the Siamese network is used to extract the features of different image pairs in the same location;considering the different contributions of different features to loop detection,the method of multi-region feature weighted aggregation is introduced to fuse the features with discrimination on the convolution feature map;then,in the convolution Geometric consistency verification is used on the feature map to determine the true loop,and the area of interest of the network is displayed in the form of a visual activation map.Finally,the experimental results on multiple public data sets show that compared with the traditional method,the algorithm proposed in this paper can It can better respond to changes in appearance and viewing angle,and can be applied to the positioning and navigation of mobile robots.