Video Stabilization using Classification-based Homography Estimation for Consumer Video Camera

Abstract

Handheld camera such as a mobile phone, action camera, and portable camcorder suffer from video instability due to unintentional shakes. This paper presents a video stabilization approach to enhance a shaky video acquired by consumer handheld camera. The proposed algorithm consists of three steps: i) generation of the binary image based on vertical class, ii) estimation of homography, and iii) frame warping for enhancing a shaky video. As a result, the proposed algorithm can remove undesired motions using optimal camera motion trajectory based on feature detection in the background region. The proposed video stabilization algorithm is suitable for consumer video devices including advanced driver assistance systems (ADAS), video surveillance systems, and aerial platforms for video enhancement.