Video Stabilization using SURF-CNN for Surveillance Application

Stabilizing videos is an essential initial stage for many computer vision tasks. Although traditional approaches have demonstrated potential outcomes, they frequently encounter difficulties when dealing with complex movements and loud surroundings. This article introduces an innovative method for video stabilization that combines Convolutional Neural Networks (CNNs) with the SURF algorithm. SURF is utilized for deriving strong characteristic points from the video frames, whereas the CNN is capable of estimating the best transformation parameters for steadying the video series. Our suggested approach surpasses current cutting-edge techniques in video stabilization quality, computational efficiency, and noise resistance, as shown by experimental results on difficult video datasets. We assess our method’s performance by utilizing quantitative metrics like mean squared error (MSE), peak signal-to-noise ratio (PSNR), and inter-frame transformation fidelity (IFT) of $\mathbf{6 8. 7 3} \mathrm{fps}$ over existing methods