Visual-Inertial Odometry Design Based on Nonlinear Optimization and Its Online Initialization Method for the Autonomous Navigation

Zhao Yibing, Wang Weiqi, Yang Yu, Xing Shuyong, Li Bin

School of Automotive engineering , Dalian University of Technology

Abstract: There is a recognized complementarity between the vision system and the inertial measurement unit (IMU) in terms of autonomous navigation.Recently, visual inertial system (VINS) has become a hotspot of current research by fusing data of low-cost inertial measurement unit and vision system.However, the lack of direct measurement information makes the estimator's initialization more difficult.This paper presents a method for online estimator initialization by using robust visual front end.One iterative process is employed to gradually align the vision system with inertial measurement unit.The convergence criterion can be used to determine the end of initialization, which can accurately recover parameters such as speed, scale, and gravity vector.This algorithm is applied to design a tightly coupled visual inertial odometry.In addition, experiments have been performed based on public data sets and equipment.The results show that the average absolute positioning error is less than 0.08m, the relative positioning error is less than 0.03m and the system has stable initialization performance with high accuracy positioning performance.

Key words: autonomous navigation,nonlinear optimization,visual inertial system (VINS), visual inertial odometer (VIO) 8w2S1hbu8as2E1ubaW1qFgFtPzQOv06icpiri2KQGXq5oFMfN+raA3dnfJ6qBlVv