一种改进的多要素视觉/惯性初始化方法  

An improved multi-element method for visual/inertial initialization

作　　者：陈润泽 郝向阳[1] 苗书锋 CHEN Runze;HAO Xiangyang;MIAO Shufeng(Information Engineering University,Zhengzhou 450001,China;Wuhan Kedao Geographical Information Engineering Co.Ltd.,Wuhan 430081,China)

机构地区：[1]信息工程大学,郑州450001 [2]武汉科岛地理信息工程有限公司,武汉430081

出　　处：《导航定位学报》2022年第2期37-43,共7页Journal of Navigation and Positioning

摘　　要：针对现有视觉/惯性初始化算法中要素不全面、估值不准确等问题,提出一种包含尺度、重力、陀螺仪零偏、加速度计偏置以及关键帧速度的多要素初始化方法,该方法通过同步估计加速度计偏置和关键帧速度来提升加速度计偏置的收敛速度和准确性,从而提升初始化的速度。此外,通过对结束时间设置了判定条件,保证在不同情况下均能快速、准确地获取初值。最后,以调整不同阶段的关键帧插入策略来提升结果的可靠性,形成一套完整的视觉/惯性初始化过程。使用尤·罗·克(EuRoC)数据集对所提算法进行验证,结果表明初值能够在多种情况下快速收敛,并且在准确度上较原方法有明显的提升,能够保证后期导航的效率。Aiming at the problems of incomplete elements and inaccurate estimation in visual/inertial initialization, this paper proposes a multi-element initialization method including scale, gravity, gyroscope bias, accelerometer bias and keyframe speed. This method improves the convergence speed and accuracy of accelerometer bias by synchronously estimating accelerometer bias and keyframe speed, which helps speed up initialization. In addition, this paper gives the judgment for the ending to ensure that initial value can be obtained quickly and accurately in different situations. Finally, by adjusting the keyframe insertion strategy in different stages to improve the reliability of the results, a complete visual/inertial initialization process is formed. In this paper, EuRoC dataset is used to verify the algorithm. The results show that initial value can converge quickly in a variety of situations, and the accuracy is significantly improved compared with the original method, which can ensure the efficiency of the later navigation.

关 键 词：视觉/惯性 初始化 关键帧 迭代 加速度计偏置 

分 类 号：P228[天文地球—大地测量学与测量工程]