一种联合文本和图像信息的行人检测方法

发布时间：2018-02-20 00:41

本文关键词： 行人检测马尔科夫随机场文本-图像信息联合共指关系指代消解　出处：《电子学报》2017年01期 　论文类型：期刊论文

【摘要】：针对纯视觉行人检测方法存在的误检、漏检率高,遮挡目标以及小尺度目标检测精度低等问题,提出一种联合文本和图像信息的行人检测方法.该方法首先利用图像分析的方法初步获取图像目标的候选框,其次通过文本分析的方法获取文本中有关图像目标的实体表达,并提出一种基于马尔科夫随机场的模型用于推断图像候选框与文本实体表达之间的共指关系(Coreference Relation),以此达到联合图像和文本信息以辅助机器视觉提高交通场景下行人检测精度的目的.在增加了图像文本描述的加州理工大学行人检测数据集上进行的测评结果表明,该方法不仅可以在图像信息的基础上联合文本信息提高交通场景中的行人检测精度,也能在文本信息的基础上联合图像信息提高文本中的指代消解(Anaphora Resolution)精度.
[Abstract]:Aiming at the problems of false detection, high missed detection rate, low precision of occluding target and small scale target detection in pure vision pedestrian detection method, A pedestrian detection method combining text and image information is proposed in this paper. Firstly, image analysis is used to obtain the candidate frame of the image target, and then the entity expression of the image object in the text is obtained by the method of text analysis. A model based on Markov random field is proposed to infer the coreference relation between image candidate box and text entity expression, so as to combine image and text information to assist machine vision to improve the downlink of traffic scene. The purpose of human detection accuracy. The result of the test on the pedestrian detection data set of California University of Technology, which has added the image and text description, shows that, This method can not only combine text information with image information to improve pedestrian detection accuracy in traffic scene, but also improve anaphora resolution accuracy on the basis of text information.
【作者单位】：武汉大学软件国家重点实验室;武汉大学计算机学院;湖南大学电气与信息工程学院;
【基金】：国家自然科学基金(No.61272274);国家自然科学基金青年项目(No.61402340) 湖北省自然科学基金(No.2014CFB194)
【分类号】：TP391.41

【相似文献】