多传感器辅助的快速图像匹配算法  

Multi-sensor Aided Fast Image Matching Algorithm

作　　者：李自豪[1] 李培林[1] 王崴[1,2] 瞿珏[1] 彭勃宇 

机构地区：[1]空军工程大学防空反导学院,陕西西安710051 [2]西安交通大学机械制造系统工程国家重点实验室,陕西西安710049

出　　处：《探测与控制学报》2016年第4期108-113,120,共7页Journal of Detection & Control

基　　金：国家自然科学基金项目资助(51405505)

摘　　要：针对现有图像匹配算法效率和精度难以兼顾的问题,提出了多传感器辅助的快速图像匹配算法。该算法利用FAST算法提取特征角点,再利用SIFT算法为特征点生成主方向和描述符;并结合基于点积的相似度度量,利用多传感器输出的姿态数据辅助搜索策略,以完成快速初匹配;最后通过统计特征点距离误差的方法剔除误匹配点,获取最终同名点集。实验表明,该图像匹配算法在效率和精度方面均优于传统的SIFT算法,能够满足增强现实系统对图像匹配算法的精度高、速度快的要求。Aiming at the problem that the existing image-matching algorithms can＇t give consideration to both efficiency and precision, a multi-sensor aided and fast matching algorithm was proposed. Firstly, the feature points were detected by using robust FAST operator. Then, the main direction of feature point and the feature point descriptors were generated by using SIFT algorithm. Afterward, the preliminary matching was finished by using the similarity matric method based on the dot product, using position and attitude data of multi-sensor to narrow the range of search. Afterwards, the final matching points set was obtained by counting the distance er-ror of feature points. Finally, the process of target tracking became more robust through template updating. Experimental results indicated that the accuracy and speed of this algorithm was superior to the traditional SIFT algorithm, which could satisfy the system requirements of higher precision and rapid speed.

关 键 词：FAST角点检测 SIFT特征 多传感器 误匹配剔除 快速匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]