Research On SLAM Algorithm Based On The Fusion Of Lidar And Vision

With the rapid development of robot technology and artificial intelligence technology,it has become a common demand for mobile robots to achieve highprecision simultaneous location and mapping in unknown environment.At present,the common SLAM methods include laser SLAM and visual SLAM.Visual SLAM can realize three-dimensional mapping.However,the RGB-D camera is sensitive to the light in the environment.Besides,it is easy to cause map drift and even mapping failure when RGB-D camera is used in areas with weak texture information.In this paper,aiming at some problems of visual SLAM when it is used alone,such as depending on feature information,sensitive to the light and only being used in static environment,a robust fusion solution for SLAM with 2D LIDAR and RGB-D camera is proposed.In this paper,firstly,a key frame extraction algorithm is designed,and the data of laser scan and RGB-D camera are collected in the mean time.According to whether the feature information in the external environment is sufficient,the switching between laser and visual SLAM mode is realized,and on this basis,a front-end odometry algorithm is designed.Next,in the consideration of the limitation of loop closure detection method based on the bag-of-words model,a novel loop closure detection method based on deep learning is proposed in this paper,and the experiment shows that this method can effectively overcome some shortcomings of bag-of-words model.Then,a back-end optimization algorithm based on semantic information is proposed to solve the problems of accuracy degradation and insufficient robustness of visual SLAM in dynamic environment.The experimental results show that this method can effectively improve the location accuracy of mobile robots in dynamic environment,and enhance the robustness of SLAM algorithm in the mean time.Finally,the point cloud filtering algorithm and octree mapping algorithm are designed.The experimental results show that our solution proposed in this paper is effective and feasible,which could be used to achieve stable location and mapping functions in the absence of visual feature or in the dynamic environment.