This paper aims to correct camera pose estimation
results in visual-inertial system (VINS) with the fish-eye camera.
Unlike existing works where optimization results are commonly
the best results, the estimation error is further reduced after
optimization. By deriving the relationship between the pose
estimation error and the map point biases, we demonstrate
the applicability of improving accuracy via map point biases.
We then propose a statistical bias calcualtion method. Our
method is integrated into a state-of-the-art VINS framework,
and the performance of the modified system is compared with
the original system on public datasets. The comparison results
confirm that our approach achieves considerable improvements
over the original system for absolute pose estimation.