| With the development of robot technology,autonomous mobile robot has been widely researched and developed.In order for a mobile robot to be able to complete tasks autonomously,it is necessary to equip the robot with navigation capabilities.For a robot to navigate in an unknown environment,it should have at least three capabilities: map building,localization,and path planning.Compared with threedimensional lidar,RGB-D cameras not only have low cost and convenient operation,but also can obtain three-dimensional environmental information.Therefore,this paper focuses on the research of map construction and path planning algorithm based on RGB-D cameras.The main contents include:(1)The visual SLAM algorithm based on RGB-D camera and the positioning method based on AMCL algorithm are studied,and the 3D map construction and positioning based on RGB-D camera are completed.Firstly,the influence of feature extraction algorithm and feature matching algorithm on the accuracy of 3D map construction in visual SLAM algorithm is analyzed.ORB algorithm is selected for feature extraction,BF algorithm for feature matching,and 3D map construction experiments are completed using data sets and in the actual environment.The results show that: Compared with SURF algorithm and SIFT algorithm,the number of feature points extracted by ORB algorithm is moderate and stable,and the algorithm takes less time.Compared with FLANN algorithm,BF algorithm can improve the accuracy of camera track.Secondly,a method is proposed to convert the point cloud map into an octree map,so as to convert the constructed three-dimensional point cloud map into an octree map,which provides a foundation for the subsequent path planning research based on octree map.Finally,the effects of the number of particles in the initial particle set and the number of random particles in the resampling stage on the AMCL algorithm were analyzed,and the optimization parameters were selected to complete the localization algorithm experiment,and the effectiveness of using the optimization parameters for localization was verified.(2)A path planning algorithm based on fusion improved A* algorithm and DWA algorithm is proposed.The path planning on a two-dimensional raster map converted from a three-dimensional point cloud map constructed by an RGB-D camera is realized.Firstly,based on the traditional A* algorithm,a critical waypoint extraction strategy based on obstacle type is proposed to eliminate redundant waypoints.The total path Angle is reduced and the global path length is shortened.The Bessel curve is used to smooth the path,and a path conforming to the robot kinematics is generated.Secondly,the improved A* algorithm is fused with the improved DWA algorithm,and the critical path points are extracted as the local target points of the improved DWA algorithm.The global path cost function is added to the evaluation function of DWA algorithm so that the fusion algorithm has both global path optimality and dynamic obstacle avoidance ability.Finally,the running experiment of the robot path planning algorithm is completed in the simulation environment and the real environment.Experimental results show that compared with the traditional A* algorithm,the improved A* algorithm reduces the path points by 52.8%,the total path Angle by5.8%,and the total global path length by 4.8%.Moreover,compared with the fusion algorithm based on the traditional A* algorithm,the efficiency of the fusion algorithm based on the improved A* algorithm is increased by 46.3%,which verifies the effectiveness of the algorithm.(3)An adaptive multi-resolution A* algorithm based on octree map is proposed to solve the problem of incomplete environment information contained in twodimensional raster maps and complete the path planning on three-dimensional maps.Firstly,based on the traditional 3D A* algorithm,an improved neighbor node search strategy is proposed.By searching six columnar neighborhoods of the current node,the search and expansion speed of the A* algorithm for adjacent nodes is improved.Secondly,based on the feature that octree maps can be used for node search under different map depths,an improved A* algorithm which can be used for adaptive multi-resolution search is proposed.By looping through the path planning at different map depths,the A* algorithm can autonomously choose the appropriate octree depth for path planning.Through the secondary path planning method to generate multiresolution path,so that the algorithm can generate path in at a faster pace cases to ensure the safety of the path.Finally,the experiment of the algorithm is completed in the data set scene and the 3D map constructed based on the RGB-D camera.Experimental results show that,compared with the traditional A* algorithm,the adaptive multi-resolution A* algorithm reduces the running time by 89.2% and the inflection point by 52.9%,which verifies the effectiveness of the proposed method and path planning in 3D octree map. |
PDF Full Text Request