Video Stabilization using point feature mapping Matlab

Abstract:

Video capturing by non-professionals will lead to unanticipated effects. Such as image distortion, image blurring etc. Hence, many researchers study such drawbacks to enhance the quality of videos. In this paper an algorithm is proposed to stabilize jittery videos. A stable output video will be attained without the effect of jitter which is caused due to shaking of handheld camera during video recording. Firstly, salient points from each frame from the input video is identified and processed followed by optimizing and stabilize the video. Optimization includes the quality of the video stabilization. This method has shown good result in terms of stabilization and it discarded distortion from the output videos recorded in different circumstances.

INTRODUCTION

This example shows how to stabilize a video that was captured from a jittery platform. One way to stabilize a video is to track a salient feature in the image and use this as an anchor point to cancel out all perturbations relative to it. This procedure, however, must be bootstrapped with knowledge of where such a salient feature lies in the first video frame. In this example, we explore a method of video stabilization that works without any such a priori knowledge. It instead automatically searches for the "background plane" in a video sequence, and uses the observed distortion to correct for camera motion.

This stabilization algorithm involves two steps. First, we determine the affine image transformations between all neighboring frames of a video sequence using the estgeotform2d function applied to point correspondences between two images. Second, we warp the video frames to achieve a stabilized video. We will use the Computer Vision Toolbox™, both for the algorithm and for display.

Step 1. Read Frames from a Movie File

Here we read in the first two frames of a video sequence. We read them as intensity images since color is not necessary for the stabilization algorithm, and because using grayscale images improves speed. Below we show both frames side by side, and we produce a red-cyan color composite to illustrate the pixel-wise difference between them. There is obviously a large vertical and horizontal offset between the two frames.

Step 2. Collect Salient Points from Each Frame

Our goal is to determine a transformation that will correct for the distortion between the two frames. We can use the estgeotform2d function for this, which will return an affine transformation. As input we must provide this function with a set of point correspondences between the two frames. To generate these correspondences, we first collect points of interest from both frames, then select likely correspondences between them.

In this step we produce these candidate points for each frame. To have the best chance that these points will have corresponding points in the other frame, we want points around salient image features such as corners. For this we use the detectFASTFeatures function, which implements one of the fastest corner detection algorithms.

The detected points from both frames are shown in the figure below. Observe how many of them cover the same image features, such as points along the tree line, the corners of the large road sign, and the corners of the cars.