Loop Closure Detection And Correction Of 3D Laser-Based SLAM With Visual Information

Simultaneous Localization and Mapping(SLAM)is one of the most important key technologies in the field of robot and automatic driving,and has very wide application prospect and physical value.Because of the characteristics and advantages of visual sensor and laser sensor,these two key technologies have become more popular in recent years.This paper firstly introduces the basic theory and method of visual SLAM and laser-based SLAM,and analyzes the advantages and disadvantages of these two methods.Relevant researches show that the lidar is more accurate in measuring data such as three-dimensional information of points,so the laser-based SLAM can reflect the information of surrounding environment well.The result maps of laser-based SLAM are quite good in small environments,but in large environments,the accumulated error becomes an important factor hindering the accuracy of the construction,and there is no effective closed loop detection method for laser SLAM to reduce the accumulated error.On the contrary,the images obtained by visual SLAM include much information and can complete the scene recognition well,but such methods still need to improve the accuracy compared with laser-based SLAM.In this paper,a method of simultaneously detecting and correcting the closed loop based on the fusion of visual information is proposed,which can solve the problem of closed loop detection of laser-based SLAM.In particular,this paper introduces a visual word bag technology used to detect and maintain visual key frames,to connect the estimated pose of the robot with the segmented laser point cloud.The method in this paper can greatly save the time cost of computing the similarity between point clouds by the fast and accurate visual closed loop detection.This paper takes the popular data set and the physical experiment as the reference and takes it as the input of the system to verify the effectiveness of the method.Finally,the experimental results on the KITTI data set and the data set collected by the author show that this method can effectively reduce the accumulated error of motion and successfully guarantee the real-time performance of closed loop correction.Compared with the original laser-based SLAM method,the optimized result of the improved algorithm presented in this paper greatly improves the accuracy and information richness,and the real-time performance.The time consumption is not too much due to the addition of visual closed-loop detection,which ensures the real-time performance of the SLAM method.The goals and achievements of this paper can be widely used in autonomous navigation robot,industrial autonomous guidance vehicle,unmanned aerial vehicle flight and other fields,which has great significance for robot map building and navigation in China.