In this paper,a new kind of pedestrian detection method is investigated;the single DMP model is combined with the couple pedestrian DPM to solve the pedestrian detection problem because of the pedestrian visual occlusion under common traffic models.This method extracts DPM feature through dataset such as INRIA,ETH,and then obtains the single/couple DPM model through LatentSVM training method.Moreover,the traffic pedestrian.distribution scene can be classified and divided first and then separated and remixed by the classification detention method.Firstly the target image will match with couple pedestrian template SDP-DPM.Secondly if couple pedestrian target can not be detected,the scene will be classi-fied as single distribution,and then matching template will switch to single pedestrian template,the results will be saved. Thirdly when the couple pedestrian are detected,the distribution will be classified and mixed,and then corresponding couple pedestrian filtering response can be saved.Finally the second matching will launch with single pedestrian template,weighted sum of the two detection results.The test proved that the method stated above can efficiently detect pedestrians under scenes that pedestrian heavily cover each other,and this method also can be more accurately compared with the traditional DPM method and other popular detection methods.%针对日常交通场景下，行人目标易被遮挡，影响行人检测效果的问题，提出一种结合单行人和双行人DPM模型的交通场景行人检测方法。该方法首先从INRIA、ETH等行人数据集中提取训练样本的DPM特征，通过La-tentSVM方法训练得到单、双人DPM模型；然后采用分类检测方法，将交通场景行人分为单独分布行人和混合分布行人两类。检测时首先使用双行人模型SDP-DPM对目标图像进行目标匹配，如果没有检测到双行人目标，则判定为单独分布行人情况，转而使用单行人模型SP-DPM进行检测，并保存检测结果；如果检测到双行人目标，则判定为混合分布行人情况，此时先保存对应的双行人滤波响应，再使用单行人模型进行二次检测，并将两次检测的结果进行加权结合。实验结果表明，本文算法能够在行人相互遮挡严重的交通环境下，有效检测出行人，整体精度优于传统的DPM算法和当前行人检测的主要流行算法。
展开▼
