摘要
针对目前单目视觉SLAM易受光照、环境、纹理影响及单目相机尺度不确定性等问题,提出一种基于视觉惯性SLAM算法VINS-Mono改进的SLAM算法。本文中算法在IMU初始化部分进行了改善,可以在更短的时间内精确地计算出陀螺仪和加速度计的偏差,并且在视觉里程计上将ORB特征点引入用来替代直接法,使其在各种复杂环境下也能准确提取特征点来跟踪相机的运动。通过在所采集真实场景与Euroc数据集下不同序列的实验与分析表明,本文中算法相较于VINS-Mono算法的鲁棒性及在定位精度上均有一定的提高,均方根误差(RMSE)平均降低17.6%。
Aiming at the problems that monocular visual SLAM is susceptible to illumination,environment,texture and the uncertainty of the scale of monocular camera,an improved SLAM algorithm based on visual inertial SLAM algorithm framework,VINS-Mono,was proposed.The algorithm in this paper is improved in the initialization part of IMU,which can accurately calculate the deviation of gyroscope and accelerometer in a shorter time.Moreover,ORB feature points are introduced into the visual otometer to replace the optical flow method,so that it can also accurately extract feature points to track camera movements in a variety of complex environments.Experiments and analyses of different sequences in real scenes and Euroc data sets show that set show that the proposed algorithm has improved robustness and accuracy of positioning with the VMs-Mono algorithm.The root mean square error(RMSE)was reduced by an average of 17.6%.
作者
俞勇杰
沈亦纯
周志峰
王立端
周围
YU Yongjie;SHEN Yichun;ZHOU Zhifeng;WANG Liduan;ZHOU Wei(School of Mechanical and Automotive Engineering,Shanghai University of Engineering Science,Shanghai 201620,China;Shanghai Satellite Research Institute,Shanghai 200240,China;ComNav Technology Ltd.,Shanghai 201801,China;Military Representative Office of the Army Armament Department Stationed in Xiangtan Prefecture,Xiangtan 411100,China)
出处
《兵器装备工程学报》
CAS
CSCD
北大核心
2024年第7期283-289,297,共8页
Journal of Ordnance Equipment Engineering
基金
上海市优秀学术/技术带头人计划资助项目(22XD1433500)。
