Research On Key Technologies Of Localization And Navigation Based On Visual SLAM

With the development of science and technology,devices with intelligent localization and navigation functions serve our lives more efficiently.Simultaneous Localization and Mapping(SLAM)technology is the foundation for localization and navigation.Laser SLAM technology is relatively mature,but the shortcomings of high price and large volume of laser sensors have limited its application in daily life.Visual SLAM has become a hot research direction in recent years.Camera sensors have low cost and rich scene content,and the real-time,accuracy and robustness of visual SLAM are gradually approaching laser SLAM.This paper mainly researches the localization and navigation of visual SLAM based on feature points.First,Aiming at the problem that the pose estimation of the visual odometer drifts with time,a visual feature pointbased localization algorithm combined with prior feature map is implemented,and innovative points are proposed to improve the localization accuracy;Second,for the problem that visual SLAM cannot directly construct a 2D occupation grid map that can be used for 2D navigation,propose an occupation grid map construction algorithm based on visual SLAM and Kinect cameras;Third,in order to solve the application problem of visual SLAM in common localization and navigation scenarios,a localization and navigation application platform based on visual SLAM was designed and established.The research work of this paper is as follows:(1)An anti-drift localization algorithm based on visual feature points is proposed.The algorithm steps include feature point extraction,2D 3D feature point matching,pose solution based on PnP,and pose optimization based on Bundle Adjust.The algorithm has two improvements.In the selection of visual feature points,the feature points trained by neural network SuperPoint are selected.Compared with the artificially designed feature points ORB,the localization accuracy is higher and the feature point performance is better;when searching for 2D 3D feature point matches The LucasKanade optical flow method was introduced at the time to improve the localization accuracy.The experimental verification shows that the algorithm meets the needs of unmanned vehicle localization in underground garages: the localization accuracy is within 10 cm,and it supports localization of data sets in different time periods.(2)A 2D occupation grid map construction algorithm based on visual SLAM and Kinect cameras is proposed.The algorithm inputs the camera trajectory after the SLAM process ends,the depth image corresponding to each camera pose,and the camera internal parameters;the algorithm outputs a 2D occupation raster map,a 3D octree map,and a 3D point cloud map.Finally,design experiments in indoor scenes verify the 2D raster map construction algorithm proposed in this chapter.(3)Design and establish a localization and navigation application platform based on visual SLAM.Based on the visual SLAM framework,the platform adds feature map saving and loading modules,occupying grid map building module,anti-drift localization modules and navigation module.Finally,the platform was applied in two common scenarios of localization and navigation,and related experimental analysis was performed.