Research And Application Of Simultaneous Localization And Mapping For Mobile Robots

Simultaneous Localization and Mapping（SLAM）technology has received extensive attention and research in the field of robot vision navigation.The SLAM technology using lidar as a sensor has problems such as high cost and being easily affected by external light,so the application scope is limited.Using cameras as sensors of SLAM technology is a major research hotspot in the field of computer vision in recent years,and vision-based SLAM has been widely used in technologies such as virtual reality and unmanned driving.This paper takes the localization and mapping of indoor vision robots as the research background,uses the depth camera Kinect-V2 as the visual sensor,and conducts in-depth learning of the four modules of SLAM technology,including sensor data processing,visual odometry,back-end optimization and map construction,mainly carried out the following work:1.An Improved ORB Algorithm with Circular Neighborhood Support and Directional Lines of Feature Points（CSD-ORB）is proposed.The Oriented FAST and Rotated BRIEF（ORB）algorithm for fast feature point extraction and description is one of the most commonly used image feature matching algorithms in visual odometry.But the ORB algorithm often has the problem of low matching point pair accuracy and no scale invariance,the feature points are also easily concentrated in the center of the image,so the image matching performance of the ORB algorithm will be reduced.Therefore,based on the classical ORB algorithm,this paper proposes a circular neighborhood support ratio to remove the wrongly matched feature point pairs,and uses the point pair direction line to uniformize the feature point distribution.Experiments on public datasets show that the CSD-ORB algorithm can better overcome the shortcomings of the classical ORB algorithm,and has good robustness,accuracy and real-time performance.2.A Loop Detection Algorithm Based On Fusion Key Frame and Center Selection Strategy（FKCS-LD）is proposed.Even with an accurate front-end visual odometer,in the process of continuous operation,the small error of the motion transformation between the two frames of images will accumulate and increase,causing the robot’s motion trajectory to drift.The loop detection algorithm in the ORB-SLAM2 system can eliminate the front-end data error and trajectory drift,but it is random and prone to frame tracking loss.In this paper,the quality of key frame selection is improved by the fusion graph algorithm,and the randomness of loop closure is reduced by the center selection strategy.The comparative experiments show that the FKCS-LD algorithm can effectively improve the efficiency and accuracy of loop detection.3.Build a software and hardware platform for mobile robots based on visual SLAM.This research work is divided into two parts: the software environment simulation and the actual field verification of the hardware platform.Part I: The complete RGB-D SLAM system improved in this paper is verified on standard public datasets,which proves that the system is real-time and accurate.Part II: Use Zhang Zhengyou’s calibration method to calibrate the Kinect V2 camera used in this paper,and use the calibrated camera,robot chassis,and embedded development board to assemble the visual SLAM robot required in this paper,and perform positioning and mapping in the indoor environment.The experiments on the software and hardware platform in this study provide a reference for the further optimization design of vision-based mobile robots.This paper studies the working principle of the visual SLAM system and the functions of each component module,proposes two algorithms for front-end visual odometry and back-end optimization,proves the effectiveness and feasibility of the research,provides a new idea for the research and development of visual SLAM.