Research On 3D SLAM Algorithm Based On Depth Camera

For robots in complex environment,using their own visual sensors to obtain the three-dimensional space model of the robot's environment and robot's trajectory,which is the work to be done by Simultaneous localization and mapping(SLAM).It is also one of the prerequisites for autonomous movement such as mission planning,navigation and control of mobile robots.RGB-D SLAM algorithm mainly consists of front-end and back-end two parts.In order to solve the problem of accumulated errors in the simultaneous localization and mapping construction of mobile robots,the traditional filtering method is applied to optimization of front-end and back-end optimization in order to improve the pose estimation and accuracy of mobile robots.The main work and achievements of this paper are as follows:1.In order to meet the real-time requirements of SLAM,this paper uses ORB algorithm to detect and describe the feature points of the image,and we adopt feature matching algorithm based on Random Sampling Consistency(RANSAC)algorithm to perform feature matching to eliminate false matches and obtain correct matching results.This not only improves the speed of image feature extraction and description,but also improves the accuracy of feature matching.2.In order to solve the exact pose of mobile robot in SLAM front end,this paper proposes an algorithm for pose solving based on EPNP combined graph optimization.First,we use EPNP to get the initial value of mobile robot's pose.Then we use the graph optimization method to optimize the pose and get a more accurate solution.By using this algorithm,more accurate pose estimation can be obtained.3.Aiming at the problem of how to detect closed loop in SLAM accurately,a closed loop detection algorithm based on bag of words is used to detect the similarity between the current key frame and all the previous key frames,so as to detect whether there is any Loopback exists.The algorithm not only improves the accuracy of loopback detection,but also speeds up the detection.4.Aiming at how to optimize the detected loopback,we use the back-end optimization method based on graph optimization to optimize the detected loopback.In the graph optimization model,only the constraint relationship between pose and pose is retained and only the pose is optimized.This approach makes the back-end optimization more concise and improves the back-end optimization efficiency.Finally,summarize the full text and put forward the prospect of further research.