Design And Development Of SLAM System For Home Service Robot Based On The Fusion Of Vision And Inertial Navigation

In the field of home service robot,SLAM is the focus of current research.The SLAM technology enables the robot to locate and build maps in an unknown environment.In reality,due to the influence of dynamic objects and other factors,the traditional SLAM is difficult to obtain more precise localization and mapping.This paper researches and designs the SLAM system of the home service robot.The main work is as follows:Firstly,an adaptive Census-SAD stereo matching algorithm is proposed in view of the disadvantages of the traditional Census algorithm and the SAD algorithm.This method uses the adaptive window Census algorithm for rough matching,then combines the adaptive weighted SAD algorithm to calculate the matching cost,and finally obtains the parallax.Compared with the traditional algorithm,this algorithm is more accurate and robust.Secondly,the dynamic feature point detection algorithm based on IMU pose estimation is designed,in view of the influence of dynamic objects on the Front End.This method obtains the matching relationship of the feature points according to the optical flow between the images,then estimates the position of the feature points in the image based on the IMU pose estimation,so as to distinguish the dynamic feature points from the static feature points.This algorithm can effectively filter out dynamic feature points,making the visual pose estimation more accurate.Thirdly,a kind of Back End optimization algorithm with the fusion of vision and IMU is designed,in the view of the cumulative error of the Front End.IMU is not affected by the environment.Vision can overcome the data drift of IMU.In this paper,the fusion can better estimate the pose of the robot.According to the graph optimization,this method uses the Damped Newton Method to reduce the visual error and IMU error,and improves the stability of the algorithm and the accuracy of the pose estimation.Fourthly,the map building for home scenes is completed.The preservation and loading of the sparse feature point map is implemented,then the dense point cloud map is built and the KNN method is used to reduce the discrete points,finally the octree map is established.Fifthly,the design and test of SLAM system is completed.The construction of the hardware platform and software system is completed.The test verifies the effectiveness of the algorithm based on public datasets and simulated home scene.