Research On Stereo Visual SLAM System With IMU

With the continuous development of artificial intelligence and sensor technology,modern human life is increasingly dependent on robots.Such as sweeping robots and combat robots,they are not only play an important role in people’s daily lives,but also reflect the national military strength.For general robots,movement is the most basic operation.In exploration and rescue missions,positioning and navigation have always been the core of its capabilities.The purpose of this article is through the fusion of IMU(Inertial measurement unit)data with stereo camera images to achieve accurate and real-time positioning function.Robot localization generally use GPS,laser radar and local area network signal strength information,in recent years,an image-based robot positioning system has been gaining popularity and is currently the only way to position robots in unknown environments.This image-based positioning method is also known as VSLAM(Visual Simultaneous Localization and Mapping)at the same time.VSLAM can be used in both indoor and outdoor environments.It overcomes the shortcomings of GPS due to weak navigation satellite signals and inaccurate positioning.Meanwhile,it can present the surrounding environment as a point cloud map,which is of great significance for the development of robots.Therefore,the main work and achievements of this paper are as follows:1.Complete the visual odometer design by using the ORB feature point method.Which mainly to optimize the extraction of feature points,through Non-maximal suppression to make feature points distribute evenly.Moreover,an adaptive distance detection method is used to complete the classification of remote and near-point of feature points,which gives higher weight to the rotation matrix obtained by calculating the near-point with small depth value,and pays more attention to the translation value of image for the far-point,thus improving the robustness of the visual odometer.2.Complete the SLAM system of vision and IMU fusion by using IMU pre-integration operation.The purpose of this paper is to integrate IMU data with visual odometer data to enhance the robustness of robot positioning system.Therefore,combining with the vins-mono system of hkust,this paper analyzes the IMU’s noise model and motion model,deduces the error propagation mode,and solves the noise propagation equation.Then,the IMU data is updated through the pre-integration operation of the IMU,and finally the visual odometer data is integrated to complete the entire positioning system.3.Using loop detection and Bag-of-Words,the vision and IMU fusion SLAM system was optimized at the back end,and the sparse 3d point cloud map was established through ORB feature points to improve the function of the whole SLAM system and the robustness of the whole system.Finally,the whole system is evaluated through the Eu Ro C test data set,and the results show that in the case of feature point loss,the system can conduct camera positioning through IMU data,achieving the purpose of real-time positioning.