Research On SLAM Method Of Indoor Mobile Robot Based On Kinect

With the improve1ent of human living standards,the demand lor indoor service robots is increasing,which makes the autonomous navigation of become an important task.The perception of unknown environment by mobile robots has always been a hotspot and diffcult point in robot intelligence.Simultaneous Localization and Mapping(SI.AM)plays a vital role,which is as one of the key technologies of Self-Localizalion and Navigation.The Visual SLAM technology is widely used because its cost-eilective and intuilive of information.Among them,the Kinect sensor has the advantage ot more direct access to three-dimensional environmental infonnation,which has been focused on research and application.With the improvement of living standards.the demand for indoor scrviec robots is increasing.which makes aulonomous navigation of become an important task.'The Poreoption of unknown environment by mobile robots has always been a hotspot and a difficult point in the research of robot intelligence.As one of the key technologies for autonomous localization and navigation,Simultaneous Localization and Mapping(SLAM)plays a vital role in development of intelligent robots.Among them,the visual sensor has the advantages of high cost performance and more intuitive information.and is widely used.This paper studies the SLAM method for the Kinect sensor that can directly obtain the three-dimensional information of the environment,and finally realizes the 3D map construction in the indoor environment.In order to improve the accuracy of visual SLAM localization the front-end visual odometry localization method was studied and improved.Accurate and fast feature matching and pose estimation are important research points for positioning accuracy.The main contents arc as follows:(1)Firstly,aiming at the problem that it is difficult to find the optimal solution in the finite iteration times of RANSAC algorithm,this paper studies a pre-screening based RANSAC matching algorithm.The improved algorithm eliminates some matching point pairs with large error by Hamming distance.and further obtains the optimal matching point pairs by double reverse matching.The experiment proves that the matching accuracy is effectively improved.(2)Secondly,the Kinect is susceptible to interference.which leads the effective matching points significantly reduced in single model estimation.In order to further improve the accuracy of localization,a motion estimation method based on hybrid model is proposed.The improved method combines the 3D-2D model to make reasonable use of the 2D feature points with missing depth information.decreasing the loss of feature matching points.And the method also combines the map information of the key frame and the previous frame,reducing the traditional method relies on adjacent frames,providing more constraint information,which greatly reduces the accumulation error and improves the localization accuracy.(3)In order to ensure the consistency of the global trajectory,loop detection based on image features is adopted,that is?the back end is combined with local and random loopback as a strategy The optimization of global pose estimation is realized by graph optimization theory.Finally,the offline datasets that provided the computer vision group of the TUM and online experiments shows the method can meet the global consistency of the trajectory.(4)Finally,the framework of Visual SLAM is designed and is validated in real scene by Turtlebot2.According to the improved localization method and the octree-based mapping,the experimental scene of three different types are realized respectively.The experiment results show that the localization method is more accurate,and the map construction based on octree structure can reflect the actual environment.In summary,this paper improves the visual SLAM algorithm,effectively improves thelocalization accuracy,and verifies the robustness of SLAM in indoor environment.At the same time,it has been applied in the field of mobile robot navigation.