电讯技术2018，Vol.58Issue(6)：682-689,8.DOI:10.3969/j.issn.1001-893x.2018.06.011
一种多视角自适应的模板匹配目标检测方法
A Multi-view Adaptive Template Matching Method for Object Detection
摘要
Abstract
Gray-level-based template matching method is difficult to solve the problem of projective varia-tions due to multiple viewpoints of images,and feature-based matching method can't yet handle the tem-plate image of small target with low resolution. Aiming at above problems, a template matching method based on optimization searching in projective transform matrix space is proposed. Firstly,the optimization search space is established by discretizing the real projective space of multiple views. Secondly,an optimi-zation model is created with the sum of absolute difference( SAD) of normalized template and real-time im-ages. Thirdly,an optimization searching algorithm is designed for finding the optimal projective matrix be-tween template and real-time images,which means that the targets in the real-time images corresponding to the template have been detected. Finally,for the high time complexity of the searching algorithm,a new algorithm based on branch and boundary is designed to accelerate the optimization searching. Experiments on VOC 2010 data-set and captured images of actual complex scene demonstrate that the proposed method performs more adaptively than SIFT-based matching method on degradations of images,such as Gaussian blurring,Gaussian noise and JPEG compression,and also more robust with lower projective error for images matching with low resolution and great difference of multiple viewpoints. Particularly,the proposed method perfectly solves matching of single template with images of multi-target.关键词
目标检测/模板匹配/多视角差异/分支界限法Key words
object detection/template matching/difference of multiple viewpoints/branch and boundary分类
信息技术与安全科学引用本文复制引用
袁伟..一种多视角自适应的模板匹配目标检测方法[J].电讯技术,2018,58(6):682-689,8.基金项目
国家自然科学基金资助项目(61374023) （61374023）
