车载大视场视觉惯性组合导航算法研究  

Research on Vehicle-mounted Large-field Visual Inertial Integrated Navigation Algorithm

作　　者：张宇 谢波 苏鑫 王萌 王猛 ZHANG Yu;XIE Bo;SU Xin;WANG Meng;WANG Meng(The 16th Institution,China Aerospace Science and Technology Corporation,Xi’an 710100)

机构地区：[1]中国航天科技集团有限公司第十六研究所,西安710100

出　　处：《导航与控制》2023年第4期34-41,4,共9页Navigation and Control

摘　　要：车载视觉导航在大角度旋转和高速运动等场景下表现不佳,由于视觉和惯性的误差特性互补,所以视觉惯性组合导航算法有良好的应用前景。大视场相机获取外界信息丰富,但大视场相机采样频率低影响视觉数据采集,而且基于优化的组合导航算法计算速度慢,故提出了一种适用于大视场相机的车载视觉惯性组合导航算法。该算法侧重于数据预处理:设计了基于伽马函数的对比度受限自适应直方图均衡化来降低图像噪声,通过FAST角点与光流法的组合完成特征提取与跟踪,进一步加快了计算速度。引入了惯性数据等效旋转矢量二子样算法来减少数据量和计算量,通过预积分算法为误差优化减少计算量,通过视觉误差、惯性误差以及边缘化误差联合优化来提高精度。与其他组合导航算法相比,该算法精度较高,计算速度最快,与VINS-Mono相比节省时间19.7%。Vehicle-mounted visual navigation performs poorly in scenarios such as large angle rotation and high-speed motion.Due to the complementary error characteristics of vision and inertia,the visual inertial integrated navigation algorithm has a good application prospect.The large field of view camera obtains rich external information,but the low sampling frequency of the camera affects the visual data acquisition.Moreover,the calculation speed of the optimized integrated navigation algorithm is slow.Therefore,a vehicle-mounted visual inertial integrated navigation algorithm suitable for large field of view camera is proposed in this paper.The algorithm focuses on data preprocessing:a contrast limited adaptive histogram equalization based on gamma function is designed to reduce image noise,and feature extraction and optical flow tracking are completed by combining FAST corner points with optical flow method to further accelerate the calculation speed.The inertial data equivalent rotation vector two-sample algorithm is introduced to reduce the amount of data and calculation,and the pre-integration algorithm is used to reduce the amount of calculation for error optimization.The accuracy is improved by joint optimization of visual error,inertial error and marginalization error.Compared with other integrated navigation algorithms,this algorithm has higher accuracy and fastest calculation speed,and saves 19.7%more time than VINS-Mono.

关 键 词：视觉惯性组合导航算法 车载导航系统 大视场相机 非线性优化 

分 类 号：V249.328[航空宇航科学与技术—飞行器设计]