复杂光照环境下的视觉惯性定位方法  

Visual inertial positioning method in complex lighting environment

作　　者：程向红[1,2] 钟志伟 刘丰宇 吴建峰[1,2] 吴昕怡 CHENG Xianghong;ZHONG Zhiwei;LIU Fengyu;WU Jianfeng;WU Xinyi(Key Laboratory of Micro-inertial Instrument and Advanced Navigation Technology,Ministry of Education,Southeast University,Nanjing 210096,China;School of Instrument Science&Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]微惯性仪表与先进导航技术教育部重点实验室,南京210096 [2]东南大学仪器科学与工程学院,南京210096

出　　处：《中国惯性技术学报》2025年第3期229-238,共10页Journal of Chinese Inertial Technology

基　　金：国家自然科学基金(62273091);国网江苏省电力有限公司省管产业单位科技项目(JC2024074)。

摘　　要：光流法假设条件严格,对光照条件、载体机动敏感。为了提高光流法特征跟踪和匹配的稳定性,提高视觉惯性定位精度,提出了一种基于精细预积分和自适应特征权重的视觉惯性定位方法。首先,在传统预积分模型的基础上,考虑惯性元件的比例因子和非正交误差,通过精细预积分得到关键帧之间的位姿变化量;其次,用其辅助光流金字塔的跟踪迭代,减少匹配搜索时间并减少特征点误匹配概率。最后,基于特征匹配置信度的差异,利用所设计的特征权重在滑窗内自适应地融合多传感器信息。实验结果表明:在EuRoC数据集中,所提方法能够有效剔除特征错误匹配;在实际实验中,相较于R-VIO、MSCKF和VINS-Mono算法,所提方法的绝对轨迹均方根误差分别平均减小了68.39%、59.06%和29.89%,证明其在各种环境下均具有较强的鲁棒性。The optical flow method assumes strict conditions and is sensitive to lighting conditions and carrier mobility.In order to improve the stability of optical flow method feature tracking and matching,and enhance the accuracy of visual inertial positioning,a visual inertial localization method based on refined pre-integration and adaptive feature weights is proposed.Firstly,based on the traditional pre-integration model,taking into account the scale factor and non-orthogonal error of the inertial sensors,the pose variation between keyframes under refined pre-integration is obtained.Secondly,using it to assist in the tracking iteration of the optical flow pyramid,which can reduce the matching search time and decrease the probability of feature point mismatches.Finally,based on the difference in confidence of feature matching,the designed feature weights are utilized to adaptively fuse multi-sensor information within the sliding window.The experimental results show that the proposed method can effectively eliminate feature mismatches in the EuRoC dataset.In actual experiments,compared with R-VIO,MSCKF and VINS-Mono algorithms,the absolute trajectory root mean square error of the proposed method is reduced by an average of 68.39%,59.06% and 29.89% respectively,indicating the robustness of the proposed algorithm to different environment.

关 键 词：视觉/惯性 光流跟踪 自适应权重 传感器融合 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]