Research On Deep Learning For Loop Closure Detection Of SLAM In Unknown Environment

With the rapid development of technology,artificial intelligence technology has been continuously improved,and robotic intelligent technology has also received widespread attention.The robot analyzes the data acquired by the vision sensor and creates an environment map in an unknown environment,and uses the map to realize robot autonomous positioning and composition.It is also called SLAM(Simultaneous Localization and Mapping)technology.With the robots in recent years,Unmanned aerial vehicles,unmanned driving,and VR/AR are popular.SLAM technology is widely known and is considered to be one of the key technologies in these fields.SLAM has less error in the case of smaller composition scale,and the local map reaches a certain level.After scale,the cumulative error of the resulting state estimates can cause serious deviations in SLAM.The loop closure detection can effectively reduce the cumulative error of SLAM.The error of loop closure detection and judgment may cause the whole SLAM map optimization algorithm to converge to the completely wrong value,the map breaks,and then the wrong positioning and navigation,so this paper focuses on the unknown environment based on loop closure detection of vision,loop closure detection is an important part of the SLAM system,and plays a very important role in eliminating cumulative errors.Although many scholars optimize around the graph optimization algorithm,loop closure detection is less robust in unknown environments.The problem is still not positively proposed.The development of deep learning provides an idea for the use of deep learning techniques to achieve loop closure detection on robots.In order to further improve the sensitivity of the loop closure detection in SLAM in practical scenarios,this paper improves the accuracy of loop closure detection through deep learning techniques,such as target detection and image segmentation,thereby improving the perception of robots in an unknown environment.The specific work is as follows:1)From the loop closure detection problem in SLAM,the reasons why the robot is sensitive in the unknown environment are analyzed,and the characteristics obtained by deep learning can better describe the scene to improve the robustness of the loop closure detection in SLAM of the robot.2)Research on the deep learning method,and choose to use depth learning to describe the image scene as the basis of this study.3)The closed-loop detection method based on static semantic region of image is studied to verify the robustness of deep learning to the loop closure detection in SLAM.4)The closed-loop detection algorithm of the robot based on the local region of interest of the image is studied to verify the robustness of deep learning to the loop closure detection in SLAM.5)Conduct experiments and analysis,summarization and outlook.