Research Of Monocular Vision SLAM Base On Point And Line Feature Detection

Simultaneous Localization and Mapping(SLAM)has become a hot topic in recent years.Researchers also use different sensors as image acquisition devices,such as monocular or binocular cameras,RGB-D cameras,inertial sensors or lidars.It is called monocular vision SLAM that only one camera is used as the only external sensor.Because of the small size and low power consumption of the camera,it is more suitable for embedded devices,which also makes the visual SLAM technology more and more attention.Therefore,the goal of this work is more challenging monocular vision SLAM.In this paper,after reviewing a large number of literatures in visual SLAM,it is impossible to accurately estimate the camera pose or even the failure condition when the feature point method monocular vision SLAM is used in the weak texture scene or motion blur.Semi-direct method monocular vision SLAM combined with dotted line feature information.In the work,the basic principle of SLAM implementation for monocular vision is given firstly,then the point-line feature detection proposed in this paper is analyzed and demonstrated in detail.Then the system framework and implementation process are given.Finally,experiments on open data sets are carried out to show the experimental results of the whole system.In this paper,the direct method and the feature method module are run in parallel by using the semi-direct method combining direct method and feature method.In the direct method module,the robustness of the direct method to the weak texture environment is utilized,the camera motion is tracked quickly and robustly,and the local semi-dense map is constructed,and the obtained information is provided to the feature method module.In the feature method module,the map is initialized according to the a priori information provided by the direct method module,the line segment feature information is added on the basis of the traditional feature points,the camera pose is jointly optimized by the point line feature,and the dotted line featureis utilized in the backend optimization.The visual dictionary tree is constructed for loop detection,which reduces the trajectory drift of the system,improves the positioning accuracy,and constructs a globally consistent feature map.In the work,the mathematical knowledge of the camera model and the geometric representation of the spatial dotted lines used in the monocular vision SLAM is first introduced.Then,the point line feature detection and matching algorithm is introduced in detail,and the camera pose is jointly optimized by the point line feature.The optimized cost function and solution process are given,and the point line detection effect diagram is shown.Then the system framework is given.The processing method of the direct method module and the feature method module is introduced in detail.The complete construction process of the dotted line feature dictionary tree is given in the feature method module.Finally,the test is performed on the public data set to verify the performance of the system.Compared with the current mainstream open source projects,the data comparison results from image registration error,absolute trajectory error and absolute attitude error.The experimental results show that the proposed method has better performance in weak texture scenes and motion blur.Robustness and accuracy...