基于改进RANSAC算法的全景图像拼接技术

现实场景中相机获取的图像视角范围往往是有限的,而实际需求又要求得到场景的全 景图,针对日常生活和工业生产中对全景图像的需求以及传统的RANSAC(random sample consensus)算法在图像配准环节因为迭代次数没有上限导致出现误匹配点对且配准 速度不高的缺陷,提出了一种改进RANSAC算法来提高全景图像拼接的效率。改进RANSAC 算法通过检测圆内的点来寻找一个最优数据检测模型,并通过粒子群算法不断更新迭代圆心 的坐标,最终得到一个最佳的匹配模型,消除特征点匹配环节出现的异常值,在提高特征 点配准的准确率的同时降低算法复杂度。在对多组图像进行拼接的实验表明,本文提出的改 进RANSAC算法相较于其他几种算法平均正确匹配率提高了9.057%, 同时算法的平均配准速率提高了5.173 s,实 现了较鲁棒的全景图像拼接效果。

Panorama image stitching technology based on improved RANSAC algorithm

In the real scene,the viewing angle range of the image obtained by the camera is often limited,and the actual demand requires the panorama of the scene.Aiming at the demand for panoramic images in daily life and industrial production.The traditional random sample consensus (RANSAC) algorithm has the defects of mismatching point pairs and low r egistration speed in image registration because there is no upper limit on the number of iteration s.This paper proposed an improved RANSAC algorithm to improve the efficiency of panoramic image stitching.The improved RANSAC algorithm finds an optimal data detection model by detecting points in the circle,and continuously updates the coordinates of the iterative circle center through the particle swarm algorithm.Finally,an optimal matching model is obtained to eliminate the outliers in the feature point matching process.It can improve the accuracy of feature point registration and reduce the complexity of the algorith m.The experiment of multi group image stitching shows that the average correct matching rate of t he improved RANSAC algorithm proposed in this paper is 9.0575% higher tha n that of other algorithms,and the average registration rate of the algorithm is 5.17375s h igher,which realizes a more robust panoramic image stitching effect.