【摘要】 针对Kinect相机存在的固有噪声,在机器人视觉定位与建图中提出一种改进ORB特征匹配算法结合改进环境测量模型的SLAM系统,该系统使用改进ORB算法提取图像的特征点,建立相邻帧之间特征点的对应关系,并对深度图进行滤波;使用ICP算法计算机器人运动,通过环境测量模型去除误匹配点,同时进行回环检测,最后使用g2o对位姿进行全局优化,建立环境点云图。通过实际环境与公开数据集的运行测试,结果表明,该V-SLAM能够准确地完成相机位姿的更新,并建立环境点云图。更多还原

【Abstract】 In the 3 D V-SLAM,due to the inherent noise of Kinect camera,a V-SLAM system combine the improved ORB features extraction algorithm with the improved environment measurement model was proposed. The improved ORB algorithm was adopted to extract the feature points of the RGB image,compute the descriptors of those points simultaneously,so that the correspondency of the feature points between the adjacent frame could be established. At meanwhile the noised of depth image were cleared. Secondly by adopting the iterative closest point algorithm,the motion estimation of the camera pose were gained. The environment measurement model could be adopted to wipe off the outlier points,refine the camera pose as well. At last,the g2 o library was used to optimize the global robot pose,construct the sparse point clouds map. According to the real scene and public dataset,the experimental result shows that the V-SLAM system can compute the transformation of the camera pose precisely,and construct the pointclouds successfully.更多还原