A robust multistage ego-motion estimation

In this paper, we assume the platform with visual observer is moved on flat ground and a constrained camera-platform configuration. The constrained camera-platform is used to simplify the camera motion model. Based on the assumption and simplified camera- platform configuration, the 3D camera motion can be separately estimated by planar region alignment between two frames. Firstly, a detected 2D motion between two frames is used to align corresponding planar regions, such an alignment or registration can removes the effects of camera rotation, and the resulting residual parallax displacement field between the two region-aligned images is an epipolar field centered at the FOE (Focus-of-Expansion). Then the camera translation can be recovered from the epipolar field. The camera rotation is recovered from the computed 3D translation and the detected 2D motion. The camera motion model is derived and represented by two parts, and a multistage ego-motion estimation process is introduced. Some experiments with real images demonstrate the robustness of the proposed motion estimation method.