Abstract: A feature-based algorithm for image mosaics of scenes with moving objects is proposed. The algorithm eliminates the feature points which are selected in moving objects using similarity transform, and the outliers are eliminated using RANSAC algorithm. In the image blend phase, a difference of two images is created using homogeneous transformation matrix, and a region growth algorithm is implemented so as to segment the moving object region. A piecewise mapping algorithm for generating panoramic images is used. Experimental results show the effectiveness of the algorithm.

Key words: image mosaics, moving objects, similarity transform, region growth, difference image

摘要： 提出一种基于特征点的运动场景图像拼接算法。基于相似变换消除选取在运动物体上的特征点，并使用RANSAC算法进行外点的剔除。在融合阶段，根据齐次变换矩阵求得对2幅图像的差异图像，并执行区域增长算法，从而分割出运动区域。使用一种分段映射算法生成全景图像。实验证明了算法的有效性。

关键词: 图像拼接, 运动物体, 相似变换, 区域生长, 差异图像

CLC Number: 

• TP391.41