【作者】 苏松志； 陈淑媛； 李绍滋；

【Author】 Song-Zhi Su1,2 , Shu-Yuan Chen3 , Shao-Zi Li1,2+ 1. Cognitive Science Department, Xiamen University, Xiamen, China, 361005, 2. Fujian Key Laboratory of the Brain-like Intelligent Systems (Xiamen University), Xiamen, China, 361005. 3. Department of Computer Science and Engineering, Yuanze University, Taiwan

【机构】 厦门大学智能科学与技术系； 福建省仿脑智能系统重点实验室(厦门大学)； 元智大学资讯工程系；

【摘要】 静态图像行人检测是计算机视觉的热点之一,行人的特征表示是研究的重点。本文针对直方图特征无法描述图像中象素点的空间位置关系这一缺点,在局部边缘模式特征的基础上,提出了结构化局部边缘矩特征。该特征不仅能够刻画局部区域的统计特性,而且能刻画边缘的局部模式及其象素点的空间分布情况。在INRIA行人数据库上的实验结果表明:采用结构化局部边缘模式矩特征的行人检测,其性能优于基于梯度方向直方图的行人检测。更多还原

【Abstract】 Pedestrian detection in static images is a challenging task in computer vision. How to represent pedestrian is the most important issue of this task. For the histogram feature cannot characterize the spatial information of local region, a novel local edge pattern based feature, structured local edge pattern moment (SLEPm), is proposed for pedestrian in the sliding window framework. SLEPm encodes both the statistical information and spatial information of object for pedestrian detection. Experimental results in INRIA pedestrian database show that the performance of SLEPm based pedestrian detection is better than that of Histogram of Oriented Gradient (HOG) based.更多还原