摘要
针对街景中的行人检测问题,提出了一种基于级联特征的行人检测方法。首先采用简单特征提取可能包含行人的潜在区域;然后采用基于概率接受的方法,结合方向梯度直方图(HOG)特征进行密集扫描分类;最后用非最大化抑制(NMS)过程聚合分类结果。实验结果表明一方面这种基于级联特征的方法使两类特征互补提高了检测精度;另一方面基于概率接受的方法进行密集扫描分类使得检测时间显著缩短。
This paper described a cascade features based algorithm towards pedestrian detection in the street scene.This method starts with using simple feature to extract potential pedestrian-included-region from the entire image.The probability receptive-based method with Histogram of Oriented Gradients(HOG) was then employed for dense scanning and classification.At last Non-Maximum Suppression(NMS) was implemented to fuse the classified results.It demonstrates in experiments that the detection accuracy is improved due to the complementation of two features from cascade features based method,and the detection time is significantly shortened for dense scanning and classification by using probability receptive-based approach.
出处
《计算机应用》
CSCD
北大核心
2011年第A02期129-132,共4页
journal of Computer Applications
基金
国家自然科学基金资助项目(60705013)
中国博士后科学基金特别资助项目(200902665)
中国博士后科学基金资助项目(20070410977)
湖南省自然科学基金资助项目(08JJ4018)
关键词
行人检测
级联特征
基于概率接受的方法
方向梯度直方图
密集扫描分类
非最大化抑制
pedestrian detection
cascade feature
probability receptive-based method
Histograms of Oriented Gradients(HOG)
dense scanning and classification
Non-Maximum Suppression(NMS)