| Popis: |
The diversity and amount of cameras are growing as these sorts of cameras are used more frequently, with people filming videos in a variety of settings. Trying to keep the camera stable and prevent shaking is not always simple, especially when using a handheld camera to record motion such as a walking tour or mountain bike ride. So far the majority of video stabilization technology has been created for recording video with a limited field of view, such as conventional videos shot with a smartphone, and it employs methods that don't translate well to 360-degree films. The architecture used by the majority of current video stabilization algorithms aid in attaining various benefits: they track gestures in the video, fit a motion model, smooth the motion, and then generate the stabilized output frames. Consequently, a feature extraction module is included in the video stabilization, and there are various ways to extract the feature. The fact that the SURF (Speeded-Up Robust-Features) is invariant to scale, rotation, translation, illumination, and blur makes them the most suitable techniques for feature detection and matching. To perform reliable estimation of inliers and outliers, hybridized RANSAC (Random sample consensus) and MSAC (M- estimator sample consensus) approaches are proposed in this work. Following this, a matched point pairs are fitted into an affine transformation model, thereby estimating the interframe motion. |
