基于先验知识类Haar特征的行人检测

Pedestrian Detection Based on Informed Haar-like Features

文章提出一种简单而实用的检测器来进行行人检测,基本思想是将人们的共识和日常的知识纳入到一种简单并且计算效率高的特征设计中去。提出的方法避免了对所有可能的长方体特征组合的穷举查找,也避免了依靠随机取样。在INRIA和Caltech行人数据集上的实验结果表明,文章的探测器在低计算量上达到了最高的水平,并且表明特征对阻塞鲁棒性非常良好。

The article propose a simple yet effective detector for pedestrian detection,its basic idea is to incorporate common sense and everyday knowledge into the design of simple and computationally efficient features. As pedestrians usually appear up-right in image or video data, the problem of pedestrian detection is considerably simpler than general purpose people detection. Experimental results on the INRIA and Cahech pedestrian datasets show that our detector reaches state-of-the-art performance at low computational costs and that our features are robust against occlusions.