中文摘要：

跨摄像机行人再识别是大型交通枢纽安防监控的基本功能,它为后续行人跨摄像机跟踪和行为识别提供支持。由于交通枢纽内行人的外观特征受遮挡、运动形变和光照变化影响显著,且目前常用的行人再识别算法对上述影响因素的鲁棒性不太理想。因此,文章提出了一种适用于大型交通枢纽的跨摄像机行人再识别算法。该算法用1个图像序列代替单幅行人图像作为查询图像,同时,采用系统抽样方法将图像序列进行分组。然后以组为单元进行相似度计算,并将计算结果作为特征训练Adaboost分类器。最后综合各分类器输出结果来判断识别结果。在iLIDS和ETHZ两个具有挑战性的数据集上进行实验,结果表明文章提出的算法优于目前其他行人再识别算法。

英文摘要：

Cross-camera pedestrian re-identification is a basic function of security monitoring in major transport hub,providing support for subsequent cross-camera tracking and pedestrian behavior recognition.As the features of the appearance of pedestrians are significantly affected by partial occlusion,motion deformation and illumination changes in transport hub,the performance of current pedestrian re-identification methods for the above influence factors is not ideal.Therefore,a cross-camera pedestrian re-identification algorithm for major transport hub was proposed,where a sequence of images was used instead of single pedestrian image as a query image,while the image sequences were grouped by a systematic sampling method.Then,the similarities were calculated by groups,and the calculated results were used as the features for AdaBoost classifier training.Last,all outputs of the classifier were compared to determine the recognition results.The experimental results on two challenging datasets,iLIDS and ETHZ,showed the superiority of the proposed algorithm to other current pedestrian re-identification algorithms.