In nature, avian animals navigate space by flapping their wings, their flapping frequency ranging from 10 Hz for larger birds to 200 Hz for small insects. Inspired by these fast oscillation-driven biological mechanisms, many scientists have designed and created numerous synthetic flapping-wing vehicles (FWVs). However, sensing autonomy using an on-board IMU is inherently challenging due to continuous large measurement perturbations on both velocity and acceleration. Learning the cycle-averaged orientation of the body plays a key role in designing a controller for the FWV. In this paper, we propose a decomposition of orientation of a hovering FWV into the averaged frame and the fast oscillating frame. By leveraging the pitching kinematics of the bird during hover, we propose a cascaded adaptive complementary filter in the special orthogonal group SO(3), where the perturbations exist entirely in a one-parameter group on SO(3), and prove the asymptotic stability of the proposed observer. The results are verified in simulation with high flapping frequency of 30 Hz, for which a conventional complementary filter with an appended geometric averaging filter fails to converge to the true average.