基于人物属性提取的行人重识别改进算法  

Improved pedestrian re⁃identification algorithm based on person attribute extraction

作　　者：梁冰锐 周卫[2] 王奔 李宏杰 杨静[3] Liang Bingrui;Zhou Wei;Wang Ben;Li Hongjie;Yang Jing(College of Electronic Information,Guangxi Minzu University,Nanning 530006,China;College of Artificial Intelligence,Guangxi Minzu University,Nanning 530006,China;College of Mathematics and Physics,Guangxi Minzu University,Nanning 530006,China)

机构地区：[1]广西民族大学电子信息学院,南宁530006 [2]广西民族大学人工智能学院,南宁530006 [3]广西民族大学数学与物理学院,南宁530006

出　　处：《现代计算机》2024年第1期9-17,共9页Modern Computer

基　　金：广西自然科学基金项目(桂科2023GXNSFBA026019)。

摘　　要：行人重识别(ReID)是指利用计算机视觉技术判断图像或者视频队列里是否存在特定行人的技术。目前,行人重识别在理论层面已经有了很好的实验结果,但在工程应用中还有很多的问题。针对在实际应用场景中行人图片内容的复杂性(摄像设备之间的差异、行人外观易受穿着、遮挡、姿态和视角等影响等)、行人识别率不高等问题,将行人重识别算法与人物属性提取算法相结合,引入注意力机制,提高了在工程应用中行人的重识别率。Pedestrian re-identification(ReID)refers to the use of computer vision techniques to determine whether specific pedestrians exist in image or video sequences.Currently,pedestrian re-identification has achieved good experimental results in theory,but there are still many challenges in engineering applications.To address the complexity of pedestrian images in practical scenarios(such as differences between cameras,influences from clothing,occlusions,poses,and viewpoints),as well as the low rec-ognition rate of pedestrians,combining pedestrian re-identification algorithms with person attribute extraction algorithms and intro-ducing attention mechanisms has improved the re-identification rate of pedestrians in engineering applications.

关 键 词：行人重识别 行人属性提取 注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]