基于ACF算法的行人检测研究  

Pedestrian detection based on ACF algorithm

作　　者：晏栋[1] 孙作雷[1] 张仓[1] 黄冠明[1] 

机构地区：[1]上海海事大学,上海201306

出　　处：《电子设计工程》2017年第17期172-175,共4页Electronic Design Engineering

摘　　要：近年来,关于行人检测领域的研究,人们提出了各种各样的方法。其中,多样化的特征和高效的分类器是这些研究的关键,本文就是在这两个方面找到一种更高效的方法来提高性能。我们的行人检测方法是将3种特征关联在一起,包括负特征(objectness features)、突出特征(salient feature)和边缘特征(edgebox feature)。与此同时,我们改进分类器的架构以提升性能。通过Caltech-USA和INRIA这两个数据库对模型进行训练,相比于一般的检测模型,我们的准确率提高了20%。In recent years, various methods in the field of pedestrian detection have been proposed. Among them, a variety of features and efficient classifier is the key to these studies, this paper is to find a more efficient way in these two areas to improve performance.Our pedestrian detection method is associated with three features, including the objectness features, salient feature and edgebox feature . At the same time, we have improved the classification architecture to improve performance.Caltech-USA and INRIA two databases train the model, compared to the general detection model, our accuracy rate increased by 20%.

关 键 词：行人检测 关联特征 边缘特征 突出特征 分类器 

分 类 号：TN99[电子电信—信号与信息处理]