摘要
针对复杂背景下的快速行人检测问题,从行人边缘信息的角度,该文提出了一种改进的中心对称统计变换(ICS_CENTRIST)特征,该特征只有32维,计算简单,描述能力强,可以很好地表达行人的边缘轮廓。行人检测时采用3级级联分类方法:第1级采用基于辅助积分图的线性支持向量机(SVM),快速排除大部分非行人区域;第2级,第3级分别使用偏最小二乘法(PLS)选出区分能力最强的前12和21个块(block),提取ICS_CENTRIST特征,采用直方图交叉核支持向量机(HIK-SVM)进行精确检测。实验结果表明,该文算法在复杂背景下可取得较好的检测效果,检测速度在447×358大小的图像上达到平均50 ms,与基于CENTRIST特征的快速检测方法和梯度方向直方图(HOG)算法相比分别提高了50%和90%,满足实时性要求。
According to the issue of pedestrian detection in complex background, this paper presents an Improved Center-Symmetric CENTRIST （ICS_CENTRIST） feature from the view of the pedestrian edge information. This feature has characters of simple calculation and powerful description ability. It can express the pedestrian＇s edge contour information perfectly with only 32 dimensions. Three cascaded classifier are used for pedestrian detection. The linear SVM based on auxiliary integral image is used for excluding most non-pedestrian area quickly in the first stage. During the second and third stages, the ICS_CENTRIST features of first 12 and 21 blocks with most strong distinguishable chosen by Partial Least Squares （PLS） method are accepted respectively, and then Histogram Intersection Kernel SVM （HIK-SVM） is used for accurate detecting. Experimental results show that this algorithm can get better detection results in complex background, and the detection speed is average 50 ms for 447 ~ 358 images, which is improved by 50% and 90% compared with the CENTRIST fast detecting and Histograms of Oriented Gradients （HOG） algorithm respectively and can meet the real-time requirements.
出处
《电子与信息学报》
EI
CSCD
北大核心
2013年第9期2040-2046,共7页
Journal of Electronics & Information Technology
基金
西南科技大学研究生创新基金(12ycjj30)
四川省教育厅基金(11ZA130)资助课题
