| Localization refers to the use of sensors to accurately determine the position of a mobile robot in an unknown environment.Obtaining accurate position information is the primary task for a mobile robot to complete map construction.Map construction refers to the robot continuously receiving sensor data in the environment and loading environmental maps to describe the occupation of objects in the environment.The algorithm framework for synchronous positioning and map construction includes: front-end and back-end optimization of visual odometry,loop detection,and map construction research.The main function of the visual odometer front-end is to extract and match feature points in adjacent two frames of images,and estimate the pose of the motion between the two frames.Backend optimization is the optimization of the pose of the visual odometer to make the results more accurate.The function of loop detection is to correct errors,helping to form a trajectory that matches the actual map after the camera detects a previously reached position.Construct a 3D map through keyframe point cloud stitching.The main research work of this article is as follows:Firstly,based on the construction of an RGB-D camera,the camera imaging principle was analyzed and an imaging model was established.The camera internal parameters were solved using Zhang Zhengyou’s camera calibration method.Secondly,estimate camera motion based on feature point method.Using the ORB feature extraction method to extract and match feature points,this paper proposes a brute force matching method combined with grid motion statistics algorithm to eliminate mismatched point pairs,leaving more interior points.Compared with the ORB+RANSAC algorithm,the interior points increase by 8.17% compared to the average.After completing the matching of feature points,the linear solution method in ICP method was used to obtain the results of camera pose changes.Then a visual bag-of-words model model is established to calculate the similarity and complete the loopback detection,which will lead to cumulative errors in the visual SLAM.The long-term estimation results are unreliable and cannot construct globally consistent trajectories and maps.Loop detection associates current data with historical data,and after tracking loss,uses loop detection for relocation.Finally,the 3D map construction was completed.On the basis of ORB-SLAM2 framework,the 3D dense point cloud map and octree map building function were added,and the experimental platform was built and run in the real laboratory scene.The 3D dense point cloud map and octree map were built.The built map can be used for subsequent navigation path planning and other work. |
