| 刘艳,武广臣.无人机影像Robert特征增强拼接算法[J].高技术通讯(中文),2025,35(10):1088~1099 |
| 无人机影像Robert特征增强拼接算法 |
| Robert feature enhanced stitching algorithm for drone images |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2025. 10. 006 |
| 中文关键词: KAZE算法； 图像信息熵； 特征匹配； 随机抽样一致性； Robert算子 |
| 英文关键词: kernel-based accelerated feature extraction algorithm, image information entropy, feature matching, random sample consensus, Robert operator |
| 基金项目: |
| 作者 | 单位 | | 刘艳 | (辽宁科技学院资源与土木工程学院本溪 117004) | | 武广臣 | |
|
| 摘要点击次数: 25 |
| 全文下载次数: 25 |
| 中文摘要: |
| 针对无人机影像拼接单应性误差积累和建筑物错位重影问题，提出一种Robert特征增强KAZE（kernel-based accelerated feature extraction）拼接算法。该算法在影像拼接前实施梯度运算，将检测边缘与原始影像灰度图进行叠加，以图像信息熵为指标筛选梯度计算最优方法。特征增强预处理后，对左右图像实施仿射 透视平移双变换后再进行拼合。对拼接图像实施渐进渐出融合处理，实现像素自然过渡和匀光匀色处理。实验结果表明，Harris、SIFT(scale-invariant feature transform)、SURF(speeded up robust features)、shi-Tomasi和KAZE这5种特征检测方法中，KAZE、SIFT和SURF算法具有相似的检测精度，然而在特征匹配方面，KAZE特征点对正确匹配率明显大于SIFT和SURF，具有较高的普适性和鲁棒性。Robert增强KAZE算法解决了建筑物拼接错位重影问题，降低了单应性误差积累，保持了图像的平均梯度，适用于大场景无人机影像连续性拼接。 |
| 英文摘要: |
| In this paper, a Robert feature enhanced kernel-based accelerated feature extraction (KAZE) stitching algorithm is proposed to address the issues of homography error accumulation and building misalignment ghosting in drone image stitching. The gradient operation is performed before image stitching, the detected edges is overlayed with the original gray image, and the information entropy is chosen to filter the optimal gradient calculation method. After feature enhancement preprocessing, the left and right images are stitched using dual transformations of left affine and right perspective translation. The concatenated images are subjected to progressive gradual out fusion processing to achieve natural pixel transitions and uniform lighting and color processing. Experimental results show that among the five feature detection methods——Harris, scale-invariant feature transform (SIFT), speeded up robust features (SURF), shi-Tomasi, and KAZE——the KAZE, SIFT, and SURF algorithms are with similar detection accuracy. However, in terms of feature matching, the correct matching rate of KAZE for feature point pair is significantly higher than that of SIFT and SURF, indicating it is with high universality and robustness. The problem of misplaced ghosting in building stitching is solved in using the Robert enhanced KAZE algorithm. It reduces the accumulation errors of homography, maintains the average gradient of the image, and is suitable for continuous stitching drone images in large scene. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
