| Simultaneous localization and mapping(SLAM)can provide a powerful solution for mobile robots and so on.It is becoming more and more important in many real-time applications and has become an active research field;Visual slam is developing towards lightweight and high precision,pursuing faster speed,higher precision and lower equipment requirements.In view of the current problems of visual Simultaneous Localization and Mapping(SLAM)for real-time applications in fast localization,and real-time indoor and outdoor dense mapping,this paper conducts a detailed analysis,then makes improvements,and finally proposes corresponding solutions.On the one hand,this paper presents ORB-SLAM2S: a complete,lightweight,accurate,and fast SLAM location solution for monocular,stereo,and RGB-D cameras,including map reuse,relocalization,and loop closing capabilities.On the other hand,this paper expands the functions of constructing dense point cloud maps and octree maps for stereo and RGB-D cameras.At the same time,for planar mobile robots,the functions of automatic external parameter calibration,dense map optimization,robust and stable ground segmentation,and 2D occupancy grid map based on plane constraints are further expanded.The above functions are all implemented under the same framework and can run in real-time on a standard Central Processing Unit(CPU).The main research content of this article consists of the following parts:A.Fast and Lightweight Localization based on Semi Optical Flow MethodIn this paper,the robust and stable feature-based method and fast optical flow method are combined to obtain the rapidity of optical flow method while ensuring the advantages of the feature-based method,which is called "semi optical flow method".At the same time,the uniform motion model is combined with the optical flow method to actively search,thereby effectively reducing the number of iterative optimizations and reducing time consumption.In this paper,ORB-SALM2 S is proposed based on ORB-SLAM2 to realize the rapid location scheme.The evaluation in popular public sequences,including the monocular,stereo,and RGB-D cameras,shows that our method achieves almost the same state-of-the-art accuracy as ORB-SLAM2 and faster speed performance which is3?5 times that of ORB-SLAM2,being in most cases the faster SLAM solution.B.An Indoor and Outdoor Dense Mapping Method for Real-time ApplicationCurrently,visual SLAM dense mapping relies on hardware acceleration such as expensive graphics processors,most of the solutions are based on RGB-D cameras that are sensitive to the outdoor environment,and the mapping scenes are limited.In response to these problems,this paper uses multi-threading and keyframe structure to dense mapping on the CPU based on fast location for indoor RGB-D and challenging outdoor stereo cameras.In this paper,based on ORB-SLAM2 S,the dense mapping scheme is extended.The evaluation in popular public sequences shows that the dense projection mapping of the RGB-D camera has achieved decimeter-level mapping accuracy.C.Dense Mapping Optimization for Planar Mobile RobotsWe observe that most of the devices that currently require dense mapping are planar mobile robots.Aiming at the unique planar characteristics of planar mobile robots,this paper further uses this feature to optimize dense maps to offset the warping caused by 6-Do F SLAM;At the same time,an automatic calibration method for the ground and camera external parameters is proposed;then,a stable ground segmentation function is realized by combining the random sampling consensus algorithm and the calibrated camera;finally,the octree map is projected to the mobile robot plane generates a twodimensional occupancy grid map that can be used for the navigation of a plane mobile robot.This paper expands the implementation of the optimization scheme based on dense mapping,and in many recognized dataset environments,through experiments,it has achieved stable and robust point cloud maps,octree maps,ground segmentation,and 2D occupancy grid map function. |
PDF Full Text Request