Research On SLAM Of Nursing Transfer Robot Based On Visual And Inertia

For indoor nursing transfer robot,accurate positioning by sensor in unknown environment is the basis of navigation and obstacle avoidance,and it is also a popular direction in the field of mobile robot research.Simultaneous Localization and map building SLAM(Simultaneous Localization and Mapping)can solve this problem perfectly.SLAM refers to the use of visual or lidar sensors to obtain ambient information to achieve their own location and the perception of the surrounding environment.For the indoor nursing transfer robot,RGB-D camera can obtain rich image information and depth information,and the positioning and mapping of the nursing transfer robot can be realized by using visual SLAM.However,the positioning may be lost in the environment with sparse features or fast movement.By integrating the data of inertial measurement unit,the problem of location loss can be solved well,and better robustness and accuracy can be obtained.Therefore,this paper mainly adopts SLAM method of RGB-D camera and inertial measurement unit fusion to carry out research.Firstly,the development history of nursing transfer robot and visual SLAM technology and the current research situation at home and abroad are investigated and analyzed,the necessity of visual and inertial integrated SLAM research is determined,and the research direction of positioning algorithm of nursing transfer robot in this subject is determined.Then,the experimental platform of nursing transfer robot was built,and the control hardware and sensor were selected considering various factors such as indoor application scenarios,space size,carrier load and volume,processor computing capacity.Then,based on visual positioning method was studied,and the key technologies for fast moving frames matching rate caused by the low,put forward the improvement method of visual SLAM visual front-end for improvement,visual front-end increased image sharpening module based on information entropy,remove the influence of the fast moving images caused by factors such as the fuzzy,and improved feature matching method is put forward.Then did an experiment with public data sets,first of all,to improve the method of feature matching experiment was carried out,and proved through the feature matching accuracy of image sharpening significantly increased,and then to visual odometer positioning precision of the experiment,the improved method and the ORB SLAM3 comparison between positioning accuracy,proves that the system robustness and accuracy of ascension.Then several experiments were carried out on the nursing transfer robot to achieve the precise positioning of the nursing transfer robot.To rely on visual positioning method of existing problems,puts forward the independent positioning method based on visual and inertial measurement unit fusion,and multiple sets of tests have been carried out in the open data set,the experimental results show that the fusion algorithm no matter from the aspects of robustness,stability and positioning accuracy are compared with the pure visual locating method has certain improvement.And it can better meet the needs of indoor positioning.Finally,the nursing displacement robot platform is used to achieve accurate positioning in real indoor environment based on the autonomous positioning method.Finally,the 3d map research of indoor nursing robot navigation is carried out.Based on the accurate pose estimation obtained by the fusion of vision and inertial navigation,the point cloud data is generated by statistical filtering and voxel filtering according to RGB-D image,and the point cloud data is processed by octree map,so as to realize 3d occupied raster map,which can be applied to robot path planning.At the same time,YOLOv5 algorithm is used to realize the detection of human standing and falling in indoor environment,and realize the online detection of human posture.