基于点转移矫正的视觉惯性里程计
Visual-inertial odometry based on point transfer correction
张梦龙 1张凯杰 1刘昌林 1周琦 1李京波1
作者信息
- 1. 华南师范大学 半导体科学与技术学院,广州 528225
- 折叠
摘要
针对卷帘相机的卷帘效应会引起图像扭曲,进而影响系统定位精度的问题,提出了一种融合IMU信息的点转移矫正的视觉惯性里程计.首先,针对卷帘图像扭曲问题,利用三焦点张量的点转移方法修正卷帘图像,相当于输入系统的是全局图像.其次,为了保证算法在嵌入式硬件上实时运行,采用扩展卡尔曼滤波进行信息融合来提高系统的定位精度和降低计算资源的要求.然后,引入静态检测和异常检测来保证系统的鲁棒性.最后,所提算法在手机上以 25 Hz左右的帧率实时运行,并在真实环境下进行实验.公开数据集上的实验结果表明:与RS-VINS-Mono算法相比,所提算法的定位精度提升了 27%,验证了该算法可有效融合视觉和惯性信息来减小卷帘效应带来的定位误差,提高了系统的鲁棒性.
Abstract
Aiming at the problem that the rolling effect of rolling shutter camera will cause image distortion and then affect the positioning accuracy of the system,a visual inertial odometry with point transfer correction based on inertial measurement unit(IMU)information is proposed.Firstly,in view of the distortion problem of the rolling image,the point transfer method of the trifocal tensor is used to correct the rolling image,which is equivalent to the global image input to the system.Secondly,in order to ensure that the algorithm runs in real time on embedded hardware,extended Kalman filter is used for information fusion to improve the positioning accuracy of the system and reduce the requirement of computing resources.Then,static detection and anomaly detection are introduced to ensure the robustness of the system.Finally,the proposed algorithm is run in real time on a mobile phone at a frame rate of about 25 Hz,and the experiment is carried out in real environment.The experimental results on the public data sets show that compared with the RS-VSIN-MONO algorithm,the proposed algorithm improves the positioning accuracy by 27%,which verifies that the proposed algorithm can effectively integrate visual and inertial information to reduce the positioning error caused by the rolling curtain effect and improve the robustness of the system.
关键词
同步定位与地图构建/视觉惯性里程计/卷帘相机/扩展卡尔曼滤波器Key words
simultaneous localization and mapping/visual-inertial odometry/rolling shutter camera/extended Kalman filter引用本文复制引用
出版年
2024
NETL
NSTL
万方数据