Research On Weak Feature SLAM Based On Binocular And Structured Light

Simultaneous Localization and Mapping(SLAM)refers to the construction of an environmental model in the process of motion without the prior knowledge of the environment,and at the same time estimating its own motion.In recent years,due to the development of visual sensors,visual SLAM has become a research hotspot.At the same time,the point feature is the most extensive and basic feature in visual SLAM.However,the extraction of point features by visual sensors has higher requirements on environmental texture.For a weak feature environment with poor texture,it is difficult for a binocular camera to calculate a depth map based on the calculation,so that the reconstructed three-dimensional image is less effective.In this thesis,binocular and structured light are combined to study SLAM in weak feature environment.The specific work is as follows:1?Binocular camera is used to collect left and right image data sets in very weak texture environment,and the data sets are pre-processed,that is,binocular correction.It mainly includes distortion correction and polar correction.Distortion correction mainly deals with the problem of image distortion caused by convex mirror.Polar line correction mainly solves the problem of incomplete level of optical axis of left and right cameras,and provides standard data sets for subsequent depth calculation.2?In binocular stereo matching,two different algorithms,semi-global stereo matching and global stereo matching,are used to obtain disparity map and depth map.At the same time,the prior knowledge based on structured light is proposed.The convolution template conforming to prior knowledge is designed to find the feature points.By the order of obtaining the feature points of the left and right images,the feature points are matched one by one,and the parallax of the feature points on the matching is calculated.The mathematical relationship model between parallax and depth is transformed into a complete SLAM reconstruction algorithm for depth maps.By comparing the effects of the depth maps obtained by the three algorithms of semi-global stereo matching,global stereo matching and structured light prior knowledge,the advantages and disadvantages of the algorithm combining the prior knowledge of structured light are verified.3?The point cloud image is generated,mainly through two different algorithms: direct linear PnP transform and minimized re-projection error,and the three-dimensional spatial transformation relationship between adjacent frames is obtained to obtain the pose.Combining the depth map and the RGB image to obtain a single-frame point cloud image,the point cloud between the adjacent frames is calculated,and the point cloud image is spliced,thereby realizing the continuous expansion of the point cloud image.