Research On Visual-Inertial Odometry Of Mobile Robot

With the popular of visual SLAM,how to make it better applied to actual mobile robots and promote its industrialization has also been one of more and more studies.Although visual odometer has achieved good results,it still has its limitations.when camera move faster in a short time,the image is blocked or image features are sparse,pure visual odometer cannot be run stably.Inertial Measurement Unit(IMU)is an internal sensor that is less disturbed by the external environment and has a high measurement frequency.It can better calculate the trajectory in a short time,but it is easy to accumulate excessive errors for a long time.So overall,the vision and IMU positioning schemes have certain complementary properties: IMU is suitable for calculating short-time and fast movements,vision is suitable for calculating long-term and slow movements.We can use the visual positioning information to estimate the zero offset of the IMU and reduce the divergence and cumulative errors caused by the zero offset of the IMU.At the same time,the IMU can also provide positioning for the rapid movement of the vision.Based on this,this paper uses nonlinear optimization and tight coupling to fuse the two positioning algorithms to form a more robust positioning algorithm,making it more suitable for industrial applications.The specific research contents are as follows:Firstly,this article introduces the importance and development history of mobile robot positioning,and also studies the development and research status of pure visual odometer and visual inertial odometer.And on this basis,the coordinate system of visual inertial odometer system based on mobile robot is constructed.Secondly,the measurement models of IMU and camera are analyzed separately,and how they estimate their own motion.According to the IMU motion model,integrated IMU data to obtain the pose.In the visual part,the fast and effective GFTT feature point plus KLT optical flow method is used for feature tracking,and the Pn P method is mainly used to replace the camera pose.Thirdly,the method of joint initialization of visual inertia is studied.The constraints between the IMU and the camera are used to estimate the initial value of the system state.The IMU pre-integration error and the inverse depth-based visual reprojection error are used to construct an optimized objective function.In order to ensure the real-time performance of the system calculation,the strategy of key frame and sliding window is used to control the calculation scale.In addition,in this paper,the marginal prior information is also added to the objective function to avoid the loss of useful information in the sliding window.Then the position and pose are obtained by minimizing the objective function,so as to realize the positioning and motion estimation of the mobile robot.Finally,the experimental verification of the visual inertial odometer solution designed in this paper is carried out,and the test comparison is first carried out on the Eu Roc data set,and then the experiment is carried out on the mobile robot.The experimental hardware platform is a mobile robot equipped with a camera and an IMU,and the experimental environment is a general indoor environment.After many different experimental tests,the validity and reliability of the visual inertial odometer designed in this paper on mobile robots are verified.Compared with pure visual odometer,this solution is more stable,more fast,and more practical.