Image moments-based velocity estimation of UAVs in GPS denied environments

This paper proposes a new method for quadrotor translational velocity estimation using only onboard sensing and without any prior knowledge on the geometry of the scene. Image moments extracted from the onboard camera images, together with onboard IMU data, are exploited in an Unscented Kalman filter (UKF). Differently with respect to the literature approaches, the proposed method is independent from the scale factor when considering a planar visual object. The estimation algorithm, in addition of involving matrices of constant size, does not require a well-textured environment, but can rather operate with few visual features. This makes it computationally-cheap, thus appealing for onboard applications. Experiments on a flying real quadrotor are carried out in a GPS-denied and poorly-textured environment in order to verify the validity of the proposed approach as well as its robustness to measurement noises and errors.