【摘要】 在分析现有的行人检测算法的基础上,针对前景提取不完整及检测误差较大等不足,提出了一种基于时空视觉显著性特征的行人检测改进算法。在具有代表性的Itti模型的基础上,使用更接近于人类视觉的Lab颜色空间对其颜色空间进行改进,并将运动特征及基于轮廓搜索的内部空洞填充法引入其中,生成总显著图。提取ROI,采用HOG特征结合SVM分类器对ROI进行行人检测。实验结果表明,该算法在一定程度上避免了误检和漏检的发生,相比较HOG算法具有较好的检效果。更多还原

【Abstract】 Through the analysis of the current pedestrian detection algorithm,for the disadvantage of incompletely extracting foreground and a large number of detecting error,a pedestrian detection algorithm based on spatio-temporal visual significant feature is proposed. On the basis of the typical Itti saliency model,use lab color space which is closer to human vision to improve the model’s color space and introduce into the motion feature and inner cavity filling method based on contour searching,generate saliency map. Extract ROI,then,using HOG feature in comnination with SVM to detect the ROI,in order to detect the pedestrian. The experimental results demonstrate that the approach can aviod some false nagative and false positive to a certain extent,and have better detecting effect compered with HOG detection algorithm.更多还原