Research On Simultaneous Localization And Mapping Based On Underwater Image Information

Simultaneous Localization and Mapping(SLAM)technology,in the unknown work area,the robot can obtain the position of the surrounding environment through its own sensor to calculate its pose,and gradually update the surrounding environment map with the positioning information.The characteristics of underwater autonomous robots are usually long-term and concealed.The visual SLAM technique can utilize visual information to provide positioning navigation when the aircraft is not equipped with a navigation system or other positioning navigation methods fail.To achieve underwater vision-based synchronous positioning and map construction solutions,the main research work of this paper is:1.Obtain camera internal parameters through image sequence information.Under the condition of only image information,the camera calibration method of this paper is different from the traditional method,and the model parameters of the camera sensor are directly obtained by using the image information,and the mathematical mapping relationship between the three-dimensional space and the two-dimensional measurement data is obtained.2.The ORB(Oriented FAST and Rotated BRIEF)feature point extraction algorithm can meet the real-time requirements of the system.For the problem that the extracted feature points are not uniform in the image,this paper uses the rasterized image sub-region extraction method.In order to improve the rate of feature point matching,this paper uses the dictionary built by the word bag model to accelerate the matching of feature points by means of motion prior knowledge.In terms of improving the correct rate of feature point matching,the statistical rule of the direction difference of matching feature point pairs is used to eliminate the false match.3.Design a visual odometer that obtains pose estimation by means of the matching relationship between the space landmark points and the image feature points.Since the monocular vision has no depth information of the feature points,triangulation is used to restore the space landmark points corresponding to the image feature points,and an initial map composed of image frames and landmark points is obtained.By optimizing the camera pose,the deviation of the spatial point from the projection point in the image to the image feature point is minimized,and the purpose of positioning the camera is achieved.4.In order to track the sustainability of the positioning,the image information of the latest moment is used to maintain a map element(keyframes and landmark points)updated with the image information.The information of the map should be representative and as concise as possible.In this paper,the design rules in the image sequence select representative key frames to join the map,and the new key frames continue to generate new landmark points with the key frames with common vision,and then the design must be Strategy to streamline redundant keyframes and landmarks.This partial map of updated keyframes and landmarks is optimized by Bundle Adjustment.5.Although visual odometers that continuously optimize the estimation results in multiple links can provide relatively satisfactory positioning results,visual odometers are still difficult to avoid the problem of cumulative errors,so a closed-loop control link is designed to correct the errors.Firstly,through the word bag model algorithm,we find the loopback keyframes similar to the current keyframes,calculate the similar changes between the two keyframes,establish the new poses of the current keyframes,and correct the keyframe poses by the pose propagation strategy.,adjust the landmark points in the map.Finally,the global positioning result is adjusted by the general graphic optimization with only the pose node.