Research On Binocular Vision SLAM Algorithm Integrating IMU

As autonomous navigation of mobile robots in an unknown environment is one of the decisive technologies to realize robot intelligence,Simultaneous Localization and Mapping(SLAM)is the core technology for their autonomous navigation.The possibilities of blurry motion images and little overlap between two frames from a pure visual SLAM may lead to failure of feature matching,system interruption and low positioning accuracy.The Inertial Measurement Unit(IMU)can provide favorable estimates for fast movements in a short period of time.Meanwhile,the camera can effectively solve the problems of static drift of IMU.Therefore,it is fair to say that IMU and camera are complementary.Monocular SLAM sees some drawbacks concerning initialization scale and tracking scale drift.This paper focuses on the algorithm of binocular visual SLAM with IMU.The research is carried out from the following aspects:(1)This research introduces the basic theoretical knowledge about binocular vision like camera pinhole model,distortion model and binocular distance measurement.By telling calibration principles of binocular cameras,IMU and binocular camera-IMU and by carrying out correspondent experiments,this paper obtains the internal and external parameters of the binocular camera,IMU parameters,external parameters of camera and IMU.(2)This paper studies the algorithm of binocular visual SLAM based on feature points.It introduces the typical binocular ORBSLAM2,including FAST and BRIEF as well as the algorithm of extracting and matching ORB feature points.It also describes the local BA optimization,loop-closure detection and overall BA optimization.Experiments of ORBSLAM2 algorithm are carried out successfully in both EUROC and Indoors recognition database.(3)This paper probes into the algorithm of binocular SLAM with IMU.The motion model at the front end is replaced by IMU data to calculate the motion between the former and the next frames.Besides,this paper directly uses the plan of VINS in the parts of back end and loop-closure detection.The modeling and pre-integration principle of IMU are introduced,and the sliding window method is used to fix the computation amount at the back end.Also,the abandoned features are added into the next optimization as priori information.Pre-integration of IMU and feature points are optimized together.The loop-closure detection is completed with DBOW.(4)Algorithms in this paper are tested in EUROC.After comparing the positioning results of algorithms in this paper with traces of open source algorithm and ground truth as well as examining deviation from open source algorithm,this paper concludes that algorithms in this paper have a profound tracking effect which can yield a continuous and complete tracking of motion with small deviations and relatively-high accuracy.