Digital Video Stabilization Method Based on Jitter Analysis of Flapping-Wing Flying Robots

Abstract

With the continuous development of flapping-wing flying robot (FWFR) technology, its applications in military reconnaissance and civil monitoring are becoming increasingly widespread. However, FWFRs experience severe image jitters in aerial videos due to their unique movement patterns. These jitters are caused by periodic wing motions that include unstable rotation along the roll axis, periodic oscillations related to wing flapping, and high-frequency mechanical vibrations. These effects significantly degrade video quality and impact subsequent visual perception tasks. To address these challenges, this paper proposes a digital video stabilization method customized for FWFRs. First, an image preprocessing module is employed to deal with the roll-axis jitters, which are caused by the factors such as robot turning and crosswind disturbance during FWFR flight. Second, in order to remove periodic and high-frequency jitters and improve the computational performance of the digital video stabilization method, we design a lightweight motion smoothing network (one learns to refine noisy motion trajectories into smooth ones) primarily comprised of stacked one-dimensional convolutional layers. Leveraging this motion smoothing network, we smooth the original motion trajectories of the video and use image warping to obtain the stabilized video. Finally, extensive video stabilization experiments under various scenarios are conducted by using our self-developed FWFR named USTB-Hawk, and results demonstrate that the proposed method achieves a PSNR of 25.15 and an SSIM of 0.761, outperforming currently employed digital video stabilization methods.