Video Stabilizing For Digital Camera

With the prevalence of handheld device like digital camera, video shot can be easily obtained. However, generally such video has the shaking problem resulted from cameraman's hands shake. Thus stabilizing process should be done to the video so as to improve the visual effect. In this paper, a framework is proposed to stabilize the shaking videos, including three modules: global motion estimation, motion filter and motion compensation. The object of global motion estimation module is to calculate the relative motion vector of adjacent frames' background, that is, the global motion vector. The most common motion styles of a camera, translation (including track and boom), roll, dolly are taken into account in our work currently. An edge-based method is introduced to deal with the luminance's sudden change between adjacent frames.Global motion includes two parts, intentional motion and camera shaking. The motion resulting from cameraman's will is intentional motion, and the difference of global motion and intentional motion is camera shaking. For smooth visual effect, motion filter should be introduced to alleviate shaking. In motion filter module, a two-order B-Spline based filter is specially designed to simulate the intentional motion and then alleviate shaking effect in the captured video. Different from the techniques before, a smooth video can be constructed simply without the prior shaking model. In the absence of blurring or distortion, camera shaking does not modify the frame content, but only shift, rotate and scale the image. The function of motion compensation module is to translate, rotate and scale the video image according to the opposition of the shaking parameters so as to achieve smooth and stable visual effect.There are two contributions in this paper: (1) In the global motion estimation module, an edge-based method is introduced to deal with the luminance's sudden change between adjacent frames; (2) In the motion filter module, a novel two-order B-Spline motion filter is proposed to alleviate shaking effect. Experimental results show that our method can suppress the video shaking and improve visual effect significantly.