基于测距信息辅助深度优化的低空飞行器单目视觉惯性里程计

A monocular visual-inertial odometry for low-altitude aircrafts with depth optimization assisted by range measurements

关键词:  视觉惯性里程计  多状态约束卡尔曼滤波器  低空飞行器  尺度恢复  测距传感器

Key words:  Visual-inertial odometry  Multi-state constrained Kalman filter  Low-altitude aircrafts  Scale recovery  Range sensors

摘要:

在卫星拒止的未知场景中，视觉惯性融合技术可作为卫星的有效替代，为飞行器提供准确的导航信息。然而在低空场景中，传统视觉惯性里程计(VIO)方案存在尺度估计不准确、容易发散及精度较低等问题。针对上述问题，提出了一种由激光测距仪/气压高度计辅助深度优化的低空飞行器单目VIO方案，以改善尺度恢复的准确性。首先，设计了基于基线约束及参考高度的鲁棒深度估计初始化方法，通过约束基线长度以提高三角化的准确性，并构建了由参考高度辅助的初始深度筛选机制以实现对深度结果的优选。随后，提出了一种基于机动状态和平面假设的深度自适应填充策略，旨在抑制尺度缺失引发的定位发散问题。最后，基于多状态约束卡尔曼滤波器(MSCKF)，融合惯性以及被测距信息修正的视觉信息对飞行器位姿进行更新。为验证所提方法，构建了搭载视觉/惯性/测距传感器的无人机试验平台，并在低空场景下开展试验验证。结果表明，与传统算法相比，所提方法定位精度平均提升61%。

Abstract:

In the unknown scenarios where satellite signals are obstructed, the visual-inertial fusion technology can serve as an effective alternative to satellites, providing accurate navigation information for aircraft. However, in the low-altitude scene, traditional visual-inertial odometry(VIO) schemes face problems such as inaccurate scale estimation, prone to divergence, and low accuracy. To address these problems, a monocular VIO for low-altitude aircrafts with depth optimization supported by laser rangefinder and barometer is proposed to improve the accuracy of scale recovery. First, a robust initialization method of depth estimation based on baseline constraint and reference altitude is designed, where the length of the baseline is constrained to improve the accuracy of triangulation, and an initial depth selection mechanism supported by reference altitude is constructed for optimal depth selection. Then, a depth adaptive filling strategy based on the aircraft maneuvering state and plane assumption is introduced to suppress the positioning divergence problem caused by scale loss. Finally, based on the multi-state constrained Kalman filter (MSCKF), the position and attitude of the aircraft is updated by integrating the inertial information and the visual information modified by range measurements. To verify the proposed method, a UAV test platform equipped with visual/inertial/range sensors is constructed to perform test verification in a low-altitude scene. The result shows that compared to the traditional algorithm, the proposed method improves the positioning accuracy by an average of 61%.