Visual Odometer And Map Construction Based On ORB Features

For the mobile carrier,which is equipped with environmental information sensor,and explores in unknown environment,locating in real-time positioning themselves,and can complete description of the environment to generate map,this process is called Simultaneous Localization and Mapping(SLAM).SLAM technology is the basic core technology of autonomous motion carrier,which is of great significance to the robot's motion planning and autonomous task completion.This article uses the depth camera as the sensor of environment perception.Its advantage is to be able to get color image and depth image information at the same time,which is well to recover true dimension of environmental information.Therefore,it has been widely used.In addition,the nonlinear optimization method based on graph optimization,it can obtain higher accuracy map and location by detecting closed loop.Based on the ORB(Oriented FAST and Rotated BRIEF)feature point,this paper constructs a SLAM system with high accuracy and robustness.The main work of this study is as follows:Firstly,the relationship between camera model and coordinate transformation is introduced,and the depth camera measurement principle and calibration method are explained in detail to prepare the input data of the system.Besides,the framework of SLAM system based on vision is presented,and the principle and data transfer mode of each module are described.Secondly,illustrate the SIFT(Scale Invariant Feature Transform)algorithm and FAST(Features from accelerated segment test)algorithm,and combine the corresponding description to give experimental verification,which is the basis for using the ORB feature point.This paper discusses the working methods of the front end visual odometers of SLAM system,and compares the PnP(Perspective-n-Point)method with the camera position trajectory based on the nonlinear optimization algorithm.In this paper,the method of bundle optimization combing space point is proposed to verify the consistency of camera trajectory.From the direct method based on visual odometer theory analysis,this paper proposes a hybrid visual odometer algorithm,which includes odometer module based on feature points and visual odometer module based on direct method.It can locate in less texture environment and has the effectiveness and superiority through experiments.In addition,the key point of this algorithm is the evaluation function,which can evaluate the odometer module position,and the evaluation feedback directly affect the choice in hybrid odometer.Finally,setting up the SLAM system,to integrate the front-end module into SLAM system,which is compared with the traditional RGB-D SLAM system,the improved system in the data set shown robust performance and higher accuracy.Besides,the backend part,unlike previous incremental building map method,this paper proposes a map creation algorithm which is based on global optimization trajectory feedback.The experimental results show that this algorithm has better map precision.Furthermore,to reduce the number of the mobile platform storage burden and also in order to be more conducive to the path planning of task,the dense point cloud map is presented in the form of octree,and experimental verification is carried out in the data set and real environment.