차량용 전방 카메라 영상을 위한 흔들림 보정 방법

Abstract

A video stabilization method for in-vehicle front camera is proposed in this paper. The proposed method generates a histogram by projecting vertical edges in the image and then estimates motions via calculating a correlation between consecutive frames. Based on motion information estimated above, the proposed system stabilizes input images. Conventional methods can cause incorrect estimation due to zooming effect. In order to reduce zooming effect, the proposed method uses the area around the vanishing point as ROI for correlation matching. Traffic signs on the road which contribute to zooming effect are removed by using color information. When big objects such as vehicles or bridges comes close to the camera, edges belong to these objects cause false correlation matching. This false matching can be avoided by verifying the value of the correlation coefficient and skipping the frame. In our experiments, The proposed method shows better results compared with conventional methods.