摘要
为了提高窗口遍历行人检测算法的分类准确率及其窗口遍历速度,提出了Fisher准则挑选特征块的行人检测算法.定尺寸梯度方向直方图特征(Histogram of Oriented Gradi-ent,HOG)块的尺寸过小,只能描述细节特征,使用变尺寸的HOG特征来描述整体与局部特征.利用Fisher准则挑选出区分能力强的特征块最终作为样本的表述符,在窗口遍历检测中利用积分图加速HOG特征的求取,使用支持向量机作为窗口分类器.实验结果表明:行人检测准确率为94%,检测时间为197 ms.
In order to increase classification precision and traversing velocity of pedestrian detection algorithm by hatch traversing,the paper presents a pedestrian detection approach to select feature block using Fisher criterion.Invariable dimension HOG feature block is so small that it can describe detailed features only.Changing dimension HOG feature block can present the overall and local features.And then,Fisher criterion is employed to select high discriminative feature to represent pedestrian.In hatch traversing integration image is used to increase the speed to get HOG feature and the vector machine is used as hatch classification.The experiment result shows that the detecting accuracy rate is 94% and detecting time is 197 ms.
出处
《西安工业大学学报》
CAS
2011年第2期109-114,共6页
Journal of Xi’an Technological University
关键词
FISHER准则
行人检测
梯度方向直方图
支持向量机
积分图
fisher criterion
pedestrian detection
histograms of oriented gradient
support vector machine
integrationi mage
