Research On Visual SLAM Method Based On Kinect Robot

Simultaneous localization and mapping(SLAM)is the key algorithm in the field of mobile robot research.The combination of the algorithm and visual sensor can give better to play its superiority.In recent years,RGB-D SLAM is represented by Kinect vision sensor is one of the important research directions in visual SLAM algorithm.In this dissertation,mobile robot visual navigation is used as the application background.Firstly,in order to solve the problem that the traditional ORB algorithm has uneven distribution of feature points and overlapping feature points in RGB-D SLAM,an improved ORB algorithm is proposed.Constructed a scale space pyramid to mesh each layer of image,increase spatial scale information;seted the region of interest,and used improved FAST corner extraction;introduced non-maximum suppression method to suppress the output of low threshold feature points.Then,aiming at resolving the shortcomings of the traditional 3D point cloud maps such as large storage space,lack of map format compactness,unclear navigational route planning,poor local space scene processing,and the three-dimensional space occupied map based on the octree data structure is applied to the RGB-D SLAM algorithm to realize the robot constructs a map of the surrounding environment.Through the simulated experiments of standard open source dataset,it is verified that the improved ORB algorithm proposed in this dissertation and octree-based data structure mapping method are feasible in RGB-D SLAM algorithm.Finally,the improved ORB algorithm and the octree-based data structure mapping method are applied to the physical verification of the RGB-D SLAM algorithm on the wheeled mobile robot independently developed by the laboratory to realize the synchronous positioning and map construction of the robot.The experimental results show that the improved ORB algorithm and the octree data structure map mode are very effective and practical as proposed and used.