曹力科, 肖晓晖. 基于卷帘快门RGB-D相机的视觉惯性SLAM方法[J]. 机器人, 2021, 43(2): 193-202. DOI: 10.13973/j.cnki.robot.200245
引用本文: 曹力科, 肖晓晖. 基于卷帘快门RGB-D相机的视觉惯性SLAM方法[J]. 机器人, 2021, 43(2): 193-202. DOI: 10.13973/j.cnki.robot.200245
CAO Like, XIAO Xiaohui. A Visual-Inertial SLAM Method Based on Rolling Shutter RGB-D Cameras[J]. ROBOT, 2021, 43(2): 193-202. DOI: 10.13973/j.cnki.robot.200245
Citation: CAO Like, XIAO Xiaohui. A Visual-Inertial SLAM Method Based on Rolling Shutter RGB-D Cameras[J]. ROBOT, 2021, 43(2): 193-202. DOI: 10.13973/j.cnki.robot.200245

基于卷帘快门RGB-D相机的视觉惯性SLAM方法

A Visual-Inertial SLAM Method Based on Rolling Shutter RGB-D Cameras

  • 摘要: 针对当前单目视觉惯性SLAM(同步定位与地图创建)中初始化需要加速度激励以及高IMU(惯性测量单元)噪声条件下系统精度下降的问题,提出一种基于卷帘快门RGB-D相机的视觉惯性SLAM方法——VINS-RSD方法.VINS-RSD方法联合卷帘快门RGB-D图像和IMU对系统进行初始化,通过控制特征的速度对卷帘快门效应进行校正,并采用一种带置信因子的损失核函数进行滑动窗口优化.为了评测该方法,在WHU-RSVI数据集的基础上制作了一个可以评价RGB-D视觉惯性SLAM算法的开源深度数据集并进行实验验证.结果表明,与VINS-Mono方法相比,VINS-RSD方法的均方根误差平均值下降了30.76%,表明该方法能获得更高的跟踪精度.

     

    Abstract: For the current monocular visual-inertial SLAM (simultaneous localization and mapping) systems, the acceleration excitation is necessary, and the system accuracy will decrease in the case of high IMU (inertial measurement unit) noises. To solve those problems, a visual-inertial SLAM method based on rolling shutter RGB-D cameras is proposed, named VINS-RSD method, which combines rolling shutter RGB-D image and IMU to initialize the system. The rolling shutter effect is corrected by controlling the feature velocity, and a loss kernel function with confidence factor is applied to the sliding window optimization. An open-source depth dataset extended from WHU-RSVI dataset is developed to evaluate the RGB-D visual-inertial SLAM method. Experiments are performed on the dataset and the root mean square error of VINS-RSD method is reduced by 30.76% compared to VINS-Mono (monocular visual-inertial system) method, which demonstrates that the proposed method can achieve a higher tracking accuracy.

     

/

返回文章
返回