Research On Key Technologies Of RGBD-SLAM In Indoor Environment

SLAM(Simultaneous Localization and Mapping)refers to the use of visual sensors to collect environmental information,and to estimate the camera motion state through the relationship between adjacent frames,and finally achieve the robot's autonomous motion in an unknown environment.Visual SLAM information acquisition generally uses visual sensors such as monocular cameras,binocular cameras and depth cameras.The depth camera can directly obtain the environmental depth information.However,the indoor environment information is complex,and there are problems such as matching accuracy and positioning accuracy.The problem has important theoretical research significance and engineering application value.In this paper,the RGBD-SLAM is studied by Kinect2.0 depth camera for indoor environment.Firstly,the RGBD-SLAM observation model based on graph optimization is constructed and optimized.On this basis,the entropy value is used to classify the acquired environment image,and a GDM(Geometry Direction Magnitude)edge descriptor is proposed for low-texture image to track camera motion.For high-texture image,ORB feature tracking camera is adopted.Then,for the uncertainty of the RGBD-SLAM system,the visual odometer uses the Random Sample Consensus(RANSAC)algorithm and the Direct-n-Point(PNP)method to estimate the camera pose.The position is further optimized by Bundle Adjustment(BA).Aiming at the cumulative error problem in RGBD-SLAM system,the loop detection method based on the word bag model is adopted,and the existing key frame screening mechanism is improved,so that the system cumulative error can be eliminated in time.Finally,experimental experiments are carried out in combination with the TUM dataset and the actual scene.The results show that the proposed GDM edge feature description algorithm performs well in low-texture scenarios.When the SLAM based on point features is not effective.Based on the combined features,the SLAM method can continue to track camera motion.At the same time,the improved key frame mechanism can effectively increase the number of loopbacks detected by the RGBD-SLAM system to reduce system uncertainty;the RGBD-SLAM system constructed in this paper can accurately locate the camera and reconstruct the environment in three dimensions.