【作者】 王喆； 修春娣；

【Author】 WANG Zhe;XIU Chundi;School of Electronic Information Engineering, Beihang University;

【机构】 北京航空航天大学电子信息工程学院；

【摘要】 针对大多数SLAM(同步定位与地图构建)方法在动态环境下定位精度与建图效果下降的问题,提出了一种基于动态目标滤除的单目视觉/IMU(惯性测量单元)融合定位方法。基于经典的视觉SLAM框架,引入语义分割对含有动态物体的图像进行动态目标滤除,进一步利用单目视觉/IMU融合定位方法实现相机的位姿估计。利用Euroc公开数据集对所提动态SLAM方法进行仿真实验,实验结果表明:动态环境下,所提算法能够有效降低绝对位姿误差和相对位姿误差,明显提升动态环境下的SLAM定位精度。更多还原

【Abstract】 Aiming at the problem that the positioning accuracy and mapping effect of most SLAM methods are reduced in dynamic environment, a monocular vision/inertial fusion localization method based on dynamic target filtering is proposed. Based on the classical visual SLAM framework, semantic segmentation is introduced to filter out dynamic objects in images containing dynamic objects, and the monocular vision/IMU fusion positioning method is further used to realize the pose estimation of cameras. The simulation experiment of the proposed dynamic SLAM method is carried out using the Euroc public dataset, and the experimental results show that the proposed algorithm can effectively reduce the absolute pose error and relative pose error in the dynamic environment, and significantly improve the SLAM positioning accuracy in the dynamic environment.更多还原