SLAM Technology Based On Laser And Vision Fusion

With the development of society and technology,people use service robots more and more frequently in their daily lives.Maps can provide information such as the location and semantics of obstacles in the home environment,allowing service robots to better understand the surrounding situation and make intelligent navigation decisions.SLAM(Simultaneous Localization and Mapping,SLAM)technology is the core technology for building maps for home environments.It can be divided into two categories according to different sensors: one is laser SLAM,which uses laser radar to obtain accurate distance information of objects on a certain plane;the other is visual SLAM,which uses cameras to obtain three-dimensional coordinates of objects within a certain angle,and combines object detection to obtain semantic information of objects.Fusion of laser and vision SLAM technology can often quickly build a more complete and semantically rich map than single-sensor mapping,but the current fusion mapping methods still have some defects: firstly,different sensors have different measurement ranges,resulting in inconsistent or incomplete map representation;secondly,the ability to handle semantic uncertainty information is insufficient,resulting in untimely or inaccurate semantic information update of the map.For these problems,this thesis proposes a SLAM technology based on laser and vision fusion with the map building demand of service robots in home environment as the application background,so as to enhance the completeness of the map and improve the accuracy of semantic information.The main work of this thesis is as follows:(1)Aiming at the problem of incomplete map building,this thesis proposes a twodimensional grid map building method based on laser and vision fusion,which expands the detection dimension of two-dimensional laser radar and realizes the perception of small obstacles below the laser radar plane at different positions,and accurately builds a grid map including environment and all obstacles.This thesis designs a pseudo-laser radar point generation method,which transforms small obstacles observed by stereo cameras into pseudo-laser radar points and fuses them with original laser radar points to expand the perception range;proposes a field-of-view differentiation method,which improves the way of grid probability update in Cartographer algorithm,so that the observation results of laser radar and stereo camera are consistent at different positions,greatly improving the completeness and robustness of map building.(2)Aiming at the problem of insufficient ability to handle semantic uncertainty information during map building,this thesis proposes a semantic grid map building method based on laser and vision fusion,which processes semantic uncertainty observed by robots from different angles and assigns more accurate semantics to occupied grids.The method first designs a semantic laser radar point generation method,which enables laser radar points to obtain accurate semantic information;then combines object size and position to design a semantic voting mechanism to process semantic uncertainty;finally processes the constructed semantic map offline and performs density clustering on grids with the same semantics to improve semantic accuracy.In summary,this thesis proposes a SLAM technology based on laser and vision fusion that enhances map usability.And by building an indoor mobile robot platform,designing realistic scenarios and comparative experiments,it verifies the completeness and robustness of grid maps and the accuracy of semantic extraction.