Abstract: The MAPMRF model is adopted to locate the Regions Of Interest(ROI). According to the imaging features of human body in infrared system, the intensity of pixels in each cirque is cumulated to construct an intensitydistance joint space based pedestrian feature representation model to hurdle the disadvantage of inertiabased feature. The pedestrian region is classified and recognized using Support Vector Machine(SVM). Experimental results on different infrared image sequences show that the proposed scheme achieves highly accurate human detection.

Key words: pedestrian detection, MAPMRF model, joint space, infrared image

摘要： 为克服亮度分布惯性特征不能充分体现人体区域亮度特征的不足，提出一种红外图像序列中的人体检测算法。采用MAPMRF模型得到人体可能存在的感兴趣区域(ROI)，根据红外图像中人体的成像特点，在以ROI中心点为圆心的各个圆环域中统计亮度信息，构建基于亮度距离联合空间的人体特征，并采用支持向量机分类器对候选区域进行分类检测。在不同红外图像序列中的实验结果均表明，该算法具有较好的鲁棒性。

关键词: 人体检测, MAPMRF模型, 联合空间, 红外图像

CLC Number: 

• TP18