欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
针对现有基于圆柱面映射的全景图像拼接算法无法实现自动估计焦距的问题，为满足实时性要求，提出了一种基于预测的快速特征点匹配算法和基于单应矩阵的焦距修正算法。该算法首先从待拼接图像中提取Harris角点，并计算方向梯度直方图（histogram of oriented gradient，HOG）描述子，采用基于预测的快速匹配算法进行特征点匹配；然后使用简化的基于纯旋转运动的焦距估计算法估计出焦距初值后，采用基于单应矩阵的焦距修正算法得到更精确的焦距值；最后将平面图像投影至圆柱平面，使用基于加权平均融合算法进行拼接，合成全景图像。采用多个图像序列进行测试得出，特征点匹配速度较传统方法提高了10倍以上，自动焦距估计算法能够准确估计摄像机焦距。拼接实验结果表明，提出的算法能够快速地合成高质量的全景图像，拼接后一图像畸变小，具有较高的实用价值。
英文摘要：
Since existing cylindrical panoramic image mosaic algorithms are unable to estimate focal length automatically, this paper presented a fast feature points matching algorithm based on prediction to meet real-time requirements, and proposed a focal length correction algorithm based on homography. Firstly, the algorithm extracted harris corners from images to be stitched, computed HOG descriptors, and performed fast feature points matching by prediction algorithm. Then it adopted a simplified method based on pure rotational motion to estimate the initial value of focal length, and obtained more accurate focus length value based on correction algorithm. Finally, it projected the plane images to cylindrical surface, and used weighted average image fusion algorithm to generate cylindrical panoramic image. It used several image test sequences to show that, the speed of feature point matching is faster than traditional method more than 10 times, and automatic focal length estimation al- gorithm can estimate focal length of the camera accurately. Stitching experimental results demonstrate that, the proposed method can synthesize high quality panoramic image rapidly and have little image distortion, and can be applied in wide applications.
同期刊论文项目
同项目期刊论文
期刊信息
位置：