基于概率分布混合模型的遮挡行人检测算法

摘要

Deformable part model achieved excellent performance in pedestrian detection,however,the score of detection would drop down to below the threshold owing to loss of detection information caused by occlusion,when handling the occlusion problem.Probability density mixture model is proposed for partly occluded pedestrian detection to increase detection score.And good results of the model are obtained by contrast experiment in INRIA,occluded INRIA and ETHZ datasets.%可变形部件模型在行人检测方面可以得到很好的检测性能,然而在处理遮挡问题时,容易出现遮挡造成检测信息的丢失,检测得分值将会下降至低于阈值而出现漏检问题.文中针对于行人检测的部分遮挡问题提出了概率密度混合模型,改善部分遮挡环境下的检测得分值,并在INRIA、遮挡INRIA以及ETHZ数据库进行对比实验,取得了良好的检测效果.