基于AKAZE的融合空间位置信息的视觉词汇树检索算法  

AKAZE-based Visual Vocabulary Tree Retrieval Algorithm Fused with Spatial Location Information

作　　者：向姗姗 陈永辉[1] 王赋攀[1] 王叶冰 XIANG Shanshan;CHEN Yonghui;WANG Fupan;WANG Yebing(School of Computer Science and Technology,Southwest University of Science and Technology,Mianyang 621010)

机构地区：[1]西南科技大学计算机科学与技术学院,绵阳621010

出　　处：《计算机与数字工程》2023年第9期1939-1944,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61802320,61872304);西南科技大学博士基金项目(编号:18zx7105,19zx7144)资助。

摘　　要：针对WebAR在自然环境中易受光照、拍照角度、拍照距离、待识别物大小等因素影响,目标识别的准确度和速度难以得到保障的问题,提出了基于AKAZE的融合空间位置信息的视觉词汇树检索方法。利用AKAZE算法提取图像局部特征,通过分层词汇树将图像局部特征量化为视觉单词,将图像检索转换为视觉单词统计匹配,量化过程融合了局部特征的空间位置信息,细化特征间的匹配重要性。以此为基础实现了一个面向图书馆和艺术馆的WebAR系统,实验结果表明,该方法能够提高自然特征识别的速度和准确率,可以进一步应用在室内外以及更复杂的交互环境中。Aiming at the problem that WebAR is susceptible to factors such as lighting,photo angle,photo distance and size of the object to be recognized in the natural environment,the accuracy and speed of target recognition are difficult to be guaranteed,a visual word tree retrieval method based on AKAZE that fuses location information is proposed.The AKAZE algorithm is used to extract image local features,quantify the image local features into visual words through a hierarchical vocabulary tree,and convert image retrieval into visual word statistical matching,the quantification process incorporates the spatial location information of local features and refines the importance of matching between features.A WebAR system for libraries and art galleries is implemented on this basis,and experimental results show that the method can improve the speed and accuracy of natural feature recognition and can be further applied to indoor and outdoor as well as more complex interactive environments.

关 键 词：WebAR AKAZE 空间位置信息 图像识别 交互 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]