基于改进SURF算法无人机影像特征匹配的研究

由于无人机受相机广角和飞行高度的限制,单张影像无法拍摄整个农田形状,导致无法准确测量农田实际面积。为此,基于图像特征匹配技术,提出改进SURF算法,用于无人机影像拼接。该算法针对传统SURF算法初始特征点选取精度不足的问题提出改进方案,优化高斯模糊的过程,进而形成新的尺度空间生成方式。通过在实验基地试验得出:本研究提出的改进SURF算法比传统SURF算法特征点在卷积核尺寸为3×3时,70 m、120 m高空的匹配率分别提升4.7%和5.3%;在卷积核尺寸为5×5时,70 m、120 m高空的匹配率分别提升4.0%和4.3%。本研究将改进后的SURF算法用于后期图像拼接中,经试验对比发现:改进的SURF算法在图片拼接处衔接程度明显提升,得到匹配精度更优的拼接图像。

Research on UAV image feature matching based on improved SURF algorithm

Due to the limitation of wide angle and flying height of the UAV, the shape of the whole farmland can not be captured by a single image, which leads to the inaccurate measurement of the actual farmland area. Therefore, based on image feature matching technology, this study proposes an improved SURF algorithm for UAV image mosaic. In order to solve the problem of inaccurate selection of initial feature points in traditional SURF algorithm, the improved algorithm optimizes the process of Gauss ambiguity, and then forms a new scale space generation method. The experimental results show that the improved SURF algorithm proposed in this paper improves the matching rate of 70 m and 120 m at high altitude by 4.7% and 5.3% respectively when the size of convolution core is 3×3 compared with the traditional SURF algorithm. When the size of convolution core is 5×5, the matching rate of 70 m and 120 m at high altitude increases by 4.0% and 4.3% respectively. In this study, the improved SURF algorithm is applied to image mosaic in the later stage. The experimental comparison shows that the improved SURF algorithm improves the degree of image mosaic significantly and achieves mosaic images with better matching accuracy.