Research On SLAM Algorithm Of Lidar And Camera Fusion In Complex Scenes

As the demands on robotic applications become increasingly complex,robots must be able to move safely,reliably,and autonomously in unknown work environments.Simultaneous Localization And Mapping(SLAM)has become one of the key technologies to achieve safe robot movement.The robot’s ability to perceive the environment also directly determines the effectiveness of the final built environment map.Due to the limitations of sensors such as lidar and depth cameras,mobile robots cannot make accurate and complete environment maps in scenes containing objects such as transparent glass and black light-absorbing panels.This thesis focuses on improving the robot’s perception capability in complex settings,and the primary research is as follows.Based on the property that lidar can only detect glass at a specific angle of incidence,this thesis adopts a simple and efficient glass detection method.First,the raw laser scan data is split into the glass and non-glass scan data by detecting the relationship between glass incidence angle and laser pulse intensity to distinguish between glass and non-glass objects.Subsequently,the glass detection algorithm is combined with the GMapping algorithm.The constructed temporary glass map matches and fuses with the original base map based on the Bayesian criterion.The improved GMapping algorithm can accurately map the glass to occupied states on the raster map.For the drawback that lidar perceives poorly when facing black panels and tends to map them incorrectly as open space,this thesis adopts a sensor fusion map building scheme.Firstly,the external parameter calibration of the lidar and the depth camera is performed to obtain the spatial transformation relationship between the lidar coordinate system and the depth camera coordinate system.Then,the pixel column compression of the depth image is converted into laser-like data to obtain the contour information of the black panel.Then the nearest value strategy is used to fuse the laser-like data with the laser data acquired by lidar,and the fused laser data is used for map construction.Finally,a SLAM map building system is proposed for complex scenes such as black panels and glass to realize SLAM map building in sets containing black panels and glass.Finally,a single lidar and depth camera based on GMapping algorithm are used to construct environment maps and experiment with SLAM building system for complex scenes such as glass and black panels,respectively.The experimental results show that the glass detection algorithm detects more than 95% of the glass area and creates an environment map containing glass information;the solution of using laser vision data fusion solves the problem of incomplete perception of black panels when using a single lidar and can build a raster map with more accurate and complete environment information.The SLAM mapping system proposed in this thesis can build complete environment maps in scenes containing glass as well as black panels,which effectively improves the safety of mobile robots moving in complex environments and extends the application scenarios of mobile robots.