RGB-DSLAM前端视觉里程计的优化算法  

RGB-DSLAM Visual Odometry Optimization Algorithm

作　　者：董静薇[1,2] 蒋燚铭 韩知雨 DONG Jing-wei;JIANG Yi-ming;HAN Zhi-yu(School of Measurement and Control Technology and Communications Engineering, Harbin University of Science and Technology, Harbin 150080, China;The Higher Educational Key Laboratory for Measuring and Control Technology and Instrumentations of Heilongjiang Province, Harbin University of Science and Technology, Harbin 150080, China)

机构地区：[1]哈尔滨理工大学测控技术与仪器黑龙江省高校重点实验室,哈尔滨150080 [2]哈尔滨理工大学测控技术与通信工程学院,哈尔滨150080

出　　处：《哈尔滨理工大学学报》2020年第6期157-164,共8页Journal of Harbin University of Science and Technology

基　　金：国家自然科学基金(61601149);黑龙江省自然科学基金(QC2017074).

摘　　要：针对视觉同步定位与地图构建领域中基于深度相机传感器的算法前端跟踪耗时长,精度差的问题,提出了一种RGB-D SLAM算法的前端视觉里程计实现方案。依据传统的视觉里程计实现方案,采用ORB特征提取算法对图像进行特征提取并计算描述子,与局部地图中地图点进行特征匹配,最后采用PnP算法计算相机位姿。为提高算法跟踪的精准度,提出在计算相机位姿后,利用三角测量进行局部地图中地图点深度的更新。实验采用TUM数据库中fr1系列数据集进行测试,结果表明,该算法在保证实时性的同时,使跟踪过程中的均方根误差平均减小了9.21%。且fr1系列两种数据集通过视觉里程计最后生成点云地图的图像信息熵比传统RGB-D SLAM系统平均提高了11.21%。In view of the long time-consuming and poor precision of RGB-D SLAM algorithm in the field of visual SLAM,an implementation scheme of RGB-D SLAM visual odometry is proposed.Based on the traditional visual odometer implementation scheme,the ORB feature extraction algorithm is used to extract the features of the image and calculate the descriptor to match the features of the map points in the local map.Finally,the PnP algorithm is used to calculate the camera pose.In order to ensure the accuracy of the tracking,it is proposed to use the triangulation algorithm to update the depth of the map points in the local map after calculating the camera pose.The experiment uses fr1 series data set in tum database to test.The experimental results show that the scheme can reduce the root-mean-square error in the tracking process by 9.21%on average while ensuring real-time performance.The image quality of the point cloud map generated is also improved by 11.21%.

关 键 词：RGB-D SLAM ORB特征提取 PnP算法 三角测量 局部地图 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]