Research On SLAM Of Robot Based On The Fusion Of LIDAR And Vision

With the improvement of technology informationization and intellectualization,the simultaneous localization and mapping of robots plays an increasingly important role in the development of mobile robots.The two most commonly used methods are visual and LIDAR methods.In the visual design,the RGB-D camera can obtain the color image and absolute depth information at the same time,which is widely used in indoor localization and mapping.However,the RGB-D camera is sensitive to light,and it has a narrow field of vision.Besides,when in the environment lack of feature,it is difficult to extract visual features for RGB-D SLAM methods.Because of the limitations above,a failed tracking may happens during practical application,which leads to a failure of building a complete map.Aiming at the above problems of localization and mapping based on RGB-D vision alone,we propose a reliable and widely applied fusion solution for SLAM with the RGB-D camera and 2d LIDAR.The main tasks of this paper include:(1)The mathematical models for two-wheeled robots,two-dimensional LIDAR and RGB-D cameras.(2)The graph optimization under the framework of RGB-D vision and LIDAR SLAM,including front-end for vision and laser,as well as the back-end optimization.The visual front-end includes feature extraction and features matching with map with the PnP algorithm.In the laser front-end,we adopt correlation matching method for laser scan matching;in the back-end,we adopt the graph optimization method.(3)To establish a real-time octomap to ensure that the laser data can be used to provide the position for the map generation even if the visual tracking fails.(4)To improve the visual SLAM with the fusion of LIDAR SLAM,including the mode switches when the visual tracking is failed and a merged localization by Extended Kalman Filter when both of visual and laser SLAM succeed.In this paper,the experiments show our solution is effective and robust,which can be used to locate and construct the image in the absence of visual feature or rapid rotation.