| 摘 要: 在全景图像拼接背景下，深入研究图像特征点的提取算法发现，ORB算法比SIFT、SURF算法的执行效率更高，但ORB算法不具备SIFT、SURF算法的尺度不变性，为了确保图像匹配的准确性，文章提出了一种动态跟踪尺度补偿的策略，对ORB算法进行优化，实现成像效果和计算效率的改善，并分析了该思路的可行性。通过OpenCV视觉库进行实验，得出提取的特征点数量约是SIFT算法的2.44 倍、SURF算法的2.24 倍、原始ORB算法的1.47 倍，验证了动态跟踪尺度补偿的ORB算法的可行性。 |
| 关键词: 动态跟踪尺度补偿；ORB算法；特征提取 |
|
中图分类号: TP311
文献标识码: A
|
| 基金项目: 广州工商学院2022 年自然科学基金项目“群体行为识别与分析技术研究”(KYYB202231). |
|
| An ORB Feature Point Extraction Algorithm based on Dynamic Tracking Scale Compensation |
|
WANG Chuanchuan, ZHOU Xueyun, WU Qibin
|
(School of Engineering, College of Technology and Business, Guangzhou 510800, China)
997459445@qq.com; zhouxy@gzgs.edu.cn; 570760876@qq.com
|
| Abstract: Based on panoramic image mosaic, in-depth study of image feature point extraction algorithm shows that ORB (Oriented FAST and Rotated BRIEF) algorithm is more efficient than SIFT and SURF algorithms, but ORB algorithm does not have the scale invariance of SIFT (Scale-invariant Feature Transform) and SURF (Speeded-Up Robust Features). In order to ensure the accuracy of image matching, this paper proposes a dynamic tracking scale compensation strategy to optimize ORB algorithm and improve imaging effect and computing efficiency. The feasibility of this idea is also analyzed. The results of the experiment with OpenCV visual library show that the number of extracted feature points is about 2.44 times of SIFT algorithm, 2.24 times of SURF algorithm, and 1.47 times of original ORB algorithm, which verifies the feasibility of ORB algorithm for dynamic tracking scale compensation. |
| Keywords: dynamic tracking scale compensation; ORB algorithm; feature extraction |
