Research On Direct Method Stereo SLAM System Fusion Of Feature Point Information

With the development of robot-related technologies,Simultaneous Localization and Mapping(SLAM)technology has attracted more and more attention from academic and industrial circles.In visual SLAM,problems such as tracking failure caused by too fast movement between camera frames and uncertain monocular SLAM dimensions are difficult points in its development.Aiming at these problems,this paper mainly studies the direct method stereo SLAM system which integrates the feature point information,and based on this,designs and implements a real-time stereo SLAM system.Finally,the system's robustness is greatly enhanced while the accuracy of the system is improved.The specific work of this article is as follows:(1)Aiming at the problem of monocular odometer scale uncertainty,a direct stereo vision odometer method is proposed which fuses feature point information.This method first uses stereo effective matching information and repeated projection for initialization.Compared with mainstream mono / stereo odometers,this method improves the efficiency and success rate of initialization and the convergence speed of map points.Second,it manages at key points.The mechanism of introducing candidate points into feature points is introduced into the structure,which shortens the overall extraction time of key points.Finally,in the sliding window optimization,stereo information constraints and geometric constraint of associated feature points on key frames are added,which not only improves the camera pose Tracking accuracy and 3D map point accuracy,and enhance the system robustness.(2)Aiming at the problem that the direct method moves too fast between frames,a direct method pose estimation method based on the fusion of feature point information is proposed.This method first determines whether the feature point information estimation module is activated.If it is activated,the candidate points are developed into feature points,and the geometric projection error of the new frame and the optimized map points in the sliding window is minimized to obtain the initial pose of the new frame.Otherwise,the direct method is used.The experimental results show that the method in this paper effectively solves the tracking failure caused by too fast motion between frames,and enhances the robustness of the system.(3)Aiming at the problem of long loop detection time caused by long-distance camera movement and the cumulative error caused by frame matching between image sequences,a fast loop detection and global optimization algorithm based on a variable search area was proposed.First,the method first determines the area to which the current key frame belongs and selects different detection methods.If the current key frame belongs to a low transmission area and a loop is detected,the area is updated,otherwise it is not updated.Experiments show that the method in this paper not only maintains the accuracy of loop detection,but also shortens the detection time to a certain extent.Finally,for the current key frame where the loop is detected,the stereo can be used to obtain the absolute scale characteristics for global optimization,speed up the global optimization,and greatly reduce the cumulative error without losing accuracy.(4)A direct method stereo SLAM system is designed and implemented which integrates feature point information.The system mainly includes four parts: camera data acquisition,data processing,direct method stereo SLAM algorithm that fuses feature information,and localization and mapping.Improved the limitation of the extraction of markers by the kalibr camera calibration system.Finally,data collection and experiments are performed on actual scenes,and compared with DSO,LDSO and other systems.Experimental results show that the system in this paper performs well in pose estimation and map construction.