VIDEO STABILIZATION USING FEATURE POINTS TRACKING

Abstract

Video Stabilization has important roll in small and compact video recording devices like mobile phones, camera, camcorders etc. Video stabilization also plays an important role in defence area. But due to camera shaking/jiggling, vibration of platform on which camera is mounted, optical turbulence, bad environmental condition, wind, moving dust particle etc., the recorded video do not have good visual quality for human operator as well as for computer vision. This report presents the robust and real time video stabilization algorithm by tracking the feature points of neighboring frames. We calculate image transformation function between neighboring frames in form of affine model. To calculate affine transformation between frames we use adaptive Random Sample Consensus (RANSAC) algorithm which is real time implementable, usable and robust to remove outliers from corresponding frames. After calculating the global motion (affine model) between neighboring frames, we remove high frequency motion to smooth out the global motion by using Gaussian low pass filter. Finally, to stabilize a video frame, we do image warping. For image warping, we transform the frames by using smooth global motion then do bilinear interpolation for resembling the frame. Proposed technique is suitable for verity of application in real time, due to very low computational cost. The wide range of applications is covered from critical defence applications to consumer electronics.