Research On Laser-SLAM Perception Method Based On Visual Semantic In Dynamic Scene

With the development of technology and economy,intelligent mobile robot has been paid more attention in recent years,the intelligent mobile robot has gradually entered people's daily life.In the application of intelligent mobile robot,the perception of environment is one of most important part of robotic function.And the speed of the robot's perception will limit the function and efficiency of the robot.SLAM(Simultaneous Localization and Mapping)is one of the intelligent mobile robotics basic function,it has been widely applied to intelligent mobile robots.Through real-time perception of information,robotic was able to location itself in the environment in real-time and building a map of.Improve the autonomy of robots by solving the problem of "where I am",and it is the prerequisite for autonomous navigation of a mobile robot.Though SLAM has been applied to real products in recent years,such as driverless vehicle,express robot,etc.However,there are still some shortcomings in the practical application of SLAM,such as the localization in dense population environment,or the change of the position and posture of the object in the environment.These situations will lead to the fact that the actual environment is not consistent with the constructed map and probably lead to the failure of SLAM location.With the development of the computer visual,we can use visual find the category and location of objects to get the visual sematic information.In order to improve the robustness of SLAM in practical applications,we will fuse the visual information and Lidar information to generate semantic laser data.By using semantic laser data,we can improve the robustness of robot localization in dynamic environment.For complete this research,the following works are done in this paper:1)Based on the study of SLAM algorithm based on laser radar,this paper analyzes the reason why SLAM has an low robustness in the dynamic environment.And proposes the use of semantic laser data to improve the robustness of the robot in the dynamic environment.2)Study the methods of visual semantic,and choose the way to meet the real-time robot perception as the basis of semantic laser data.3)The clustering algorithm of laser data is studied,and the clustering algorithm is improved according to the actual application.4)In order to generate real-time semantic laser data,the fusion method of visual semantic information and laser scan information.5)In order to verify the robustness of semantic laser information data to robot localization,a SLAM localization method experiment based on semantic laser data is carried out.