VINS-mono Optimized: A Monocular Visual-inertial State Estimator with Improved Initialization
Abstract
State estimation is one of the key areas in robotics. It touches a variety of applications in practice such as, aerial vehicle navigation, autonomous driving, augmented reality, and virtual reality. A monocular visual-inertial system (VINS) is one of the popular trends in solving state estimation. By fusing a monocular camera and IMU properly, the system is capable of providing the position and orientation of a vehicle and recovering the scale.
One of the challenges for a monocular VINS is estimator initialization due to the inadequacy of direct distance measurement. Based on the work of Hong Kong University of Technology on monocular VINS, a checkerboard pattern is introduced to improve the original initialization process. The checkerboard parameters are used along with the calculated 3D coordinates to replace the original initialization process, leading to higher accuracy. The results demonstrated lowered cross track error and final drift, compared with the original approach.
Citation
Xu, Lingjie (2018). VINS-mono Optimized: A Monocular Visual-inertial State Estimator with Improved Initialization. Master's thesis, Texas A & M University. Available electronically from https : / /hdl .handle .net /1969 .1 /174383.