Onboard optical flow and vision based localization for a quadrotor in unstructured indoor environments

Abstract

This paper considers problem of localization for an aerial robot in unstructured indoor environments. A vision correction method based on linear characteristics is proposed. The dynamic model is established using an optical flow sensor. Use a strap-down camera to capture the frame of perpendicular line on the ground, and calculate the global position of their intersection point. As an observation, the vision position is input to the KALMAN filter in order to eliminate the accumulated error. Experimental results from indoor hover test verified the accuracy and real-time of the approach.