3D video stabilization for a humanoid robot using point feature trajectory smoothing

This paper proposes a new 3D video stabilization method for a humanoid robot without explicit global camera motion estimation with respect to a reference frame. 2D video stabilization that uses a 2D camera motion model works well if scenes are far from the camera of the robot or can be modeled as a plane. However, because a humanoid robot operates in a real 3D space, these 2D video stabilization approaches may fail to stabilize unstable videos. Furthermore, straightforward 3D video stabilization methods that depend on structure-from-motion to estimate global camera motion might fail in the long run because error accumulation in the global motion estimation step is inevitable with the elapse of time. Instead, our method uses only the relative 3D camera motion between the real unstable camera and the virtual stabilized camera through point feature trajectory smoothing. In order to evaluate our method, we use a real humanoid robot that is able to walk and run. The experimental results show that the proposed 3D video stabilization system is a potential candidate for application in humanoid robots.