VO estimates the rough camera movement based on the information of the consecutive images, and provides a good initial value of the back end.
VO are mainly divided into two categories: feature point method and direct method.