基于Harris算法的无人机影像拼接技术
2018-05-15赵文君李枭
赵文君 李枭

摘 要:处理无人机影像数据时,Harris角点检测算法具有较强的鲁棒性和稳定性。使用Harris角点检测算法时,影像边界处由于影像畸变影响,生成的特征点存在角点聚簇和伪角点的概率非常大,在处理该类问题时,通常是删除影像边界生成的角点。针对影像边缘特征点删除量的合理性进行了实验:先对影像生成特征点面积进行限制,用Harris算法提取特征点。然后用非极大抑制算法选取特征点,去除伪角点和聚簇的角点,生成最佳缝合线进行影像匹配融合。最后对比影像生成特征点面积和最后影像匹配效果,论证影像边缘删除Harris角点的合理量。实验结果表明,相对于传统的直接删除边界Harris角点方法,该方法更可靠更精确。
关键词:Harris算法;影像拼接;无人机影像
DOI:10.11907/rjdk.172721
中图分类号:TP317.4
文献标识码:A 文章编号:1672-7800(2018)004-0216-04
Abstract:When processing the data of Unmanned Aerial Vehicle (UAV) image, Harris corner detection algorithm has strong robustness and stability. Because of image distortion in its boundaries with Harris corner detection algorithm, there is a very large probability for characteristic points to be corner clusters and pseudo corner points, so when processing this problem, the corner points generated by the image boundaries are usually removed. In order to extract the UAV image characteristic points by Harris algorithm, the rationality of the deletion of the characteristic points which are generated in the image boundaries is studied,this paper launches the experiment at this point.The area of image-generated characteristic points is limited first, and the extraction of characteristic points is carried on by Harris corner detection algorithm next, and the characteristic points selected by non maximum suppression algorithm are used to remove the pseudo corner points and corner clusters and generate the best suture to match and fuse images; finally, having comparing the relationship between the area of image-generated characteristic points and the final image matching effect, the reasonable number of deleted Harris corners in the image boundaries is demonstrated. The experimental results show that the method of this paper is more accurate and reliable than the traditional method of deleting Harris corners of the boundaries directly.
Key Words:Harris algorithm; image stitching; UAV image
0 引言
无人机航测节省了大量人力物力,使影像获取更加便捷。但无人机体积小、质量轻,在飞行时容易受到气流影响,拍摄出来的照片存在畸变,影响后续影像拼接。
影像拼接技术指将一组重叠影像集合拼接成一幅大范围的无缝高分辨率影像[1]。影像拼接技术包括角点提取、图像配准和图像融合3个关键步骤[2]。已有的图像匹配算法主要有基于特征点的配准、基于灰度信息的配准以及基于变换域的配准3大类。基于特征点配准算法可对图像中特征稳定的点进行匹配,减少了信息处理量,对图像的灰度变化具有鲁棒性[3]。目前提取影像特征点的主流算法有SIFT算法[1]、Moravec算法[4]、Harris算法[5]、SUSAN算法[6]等,本文选用Harris角点检测算法。……
