Image Stitching Method for Surround View Image without Seamline

Jinsol Choi, Heunseung Lim, Sangwoo Yun, Minwoo Shin, Joonki Paik

Abstract: The surround view system delivers the surrounding environment to the driver in a top-down manner so that the driver can recognize blind spots. In this system, image stitching is combining camera images facing different directions into a single image. As an easy-to-access method, there is a method of warping through homography by extracting and matching the features of the images to be stitched. However, this method has a limitation in that stitching is not performed properly if the geometric distortion is severe. In order to solve this problem, a method of estimating a seam and stitching images is being studied in many existing methods, and a representative method is a method using a distance matrix using the difference of gradient components. However, this method has a problem of falling into the local minima and causing a loop phenomenon. To solve this problem, we propose a seamline estimation method using ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</inf> -norm based gradient priors. The optimal seam can be found by obtaining the energy matrix through ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</inf> -gradient priors and estimating the distance matrix through this. In this method, improved seam estimation is possible because the shortest distance is calculated as the point where the rate of change between the two images is minimized.