UAV image stabilization algorithm based on improved Shi-Tomasi

As the UAV is susceptible to its own engine vibration and irregular airflow in the inspection process, the image sequence collected has jitter, which affects the reliability of the inspection results. To solve this problem, this paper proposes an improved Shi-Tomasi algorithm combined with a selective mechanistic motion model for UAV inspection image stabilization algorithm to eliminate the jitter component of images and produce stable images. Firstly, the feature points in the image are extracted by using the improved Shi-Tomasi algorithm, and feature point matching is performed by using optical flow tracking. Then the motion model is determined to describe the inter-frame motion by a selection mechanism based on the matched feature point pairs. The obtained original motion trajectories are turned into smooth trajectories by sliding filtering, and finally stable images are generated by warping and cropping. The experimental data use the UAV inspection image sequences collected by ourselves. The experimental results demonstrate the good performance of the proposed algorithm, which can quickly and effectively eliminate the image dithering.