基于加权预积分和快速初始化的惯性辅助单目前端模型

An inertial assisted monocular front-end model based on weighted pre-integration and fast initialization

针对单目视觉惯性定位系统在复杂环境和相机高动态条件下的实时性和高精度的需求,提出了一种基于加权预积分和快速初始化的惯性辅助单目前端模型Improved_VIO。首先同步视觉和惯性测量数据,建立高精度的IMU加权预积分模型,为联合初始化和视觉跟踪模型提供帧间运动约束;然后构建视觉惯性融合状态向量,建立联合初始化模型,实现视觉惯性松耦合的快速联合初始化;最后在IMU加权预积分和快速初始化方法的基础上,建立一套惯性辅助的视觉跟踪模型,从而有效提高系统定位精度。在EuRoC数据集上的试验结果表明,与传统视觉惯性定位前端模型相比,本文的前端模型提升了单目视觉惯性定位的精度与实时性,初始化时间缩短至10 s内,定位精度提高了约30%。

Aiming at the robustness and high precision of monocular visual inertial positioning system in complex environment and camera dynamic conditions, Improved_VIO, an inertial assisted monocular front-end model based on weighted pre-integration and fast initialization is proposed. Firstly, the visual and inertial measurement data are synchronized, and a high-precision IMU weighted pre-integration model is established to provide inter-frame motion constraints for joint initialization and visual tracking models. Secondly, constructing the visual inertia fusion state vector, and establishing the joint initialization model realize the fast joint initialization of visual inertia coupling. Finally, based on the IMU weighted pre-integration and fast initialization methods, a visual inertia-assisted tracking model is established to effectively improve the robustness of the system. The experimental results show that compared with the traditional visual inertial positioning front-end model, the Improved_VIO improves the accuracy, speed and robustness of monocular visual inertial positioning. The initialization time is shortened to 10 seconds, and the positioning accuracy is improved about 30%.