张振亚,屈兴亮,王萍,方博,孙兆隽.基于步态识别的建筑物内多人员目标辨识方法研究[J].安徽建筑大学学报,2023,31(4):70-76 |
基于步态识别的建筑物内多人员目标辨识方法研究 |
Research on Multi-Occupant Identification Method in Building Based on Gait Recognition |
|
DOI: |
中文关键词: 建筑智能化 步态分割 步态识别 |
英文关键词: building intellectualization gait segmentation gait recognition |
基金项目:安徽省高校省级自然科学研究项目(KJ2020A0470);安徽省高校学科拔尖人才学术资助项目(gxyq2022030、gxbjZD2021067) |
|
摘要点击次数: 4001 |
全文下载次数: 0 |
中文摘要: |
人员的身份辨识及其时空分布是建筑智能化中的基础输入信息。针对建筑物内多人员目标同时辨识的问题,本文设计了基于步态识别的建筑物内多人员辨识(multi-occupant identification based on GaitSet,MOIG)框架。对于通过摄像头采集的建筑物内视频流,该框架首先利用 YOLOv5 和 DeepSort 构建视频中出现的多个人员步态序列,然后利用 DeepLabv3+ 分割模型分割人员步态序列图像,再利用 GaitSet 辨识人员身份。实验表明,该框架在真实场景下的人员辨识准确率可达 75%,与常规方法相比,性能提升显著。 |
英文摘要: |
The occupant identification and spatio-temporal distribution is the basic input information required in buildingintellectualization. To identify multiple occupants in buildings simultaneously,a multi-occupant identification based on GaitSet(MOIG)is proposed. The framework collects the video streams inside the building captured by the cameras,constructs multiple gait sequences ofpeople appearing in the video with YOLOv5 and DeepSort,segments the gait sequence images by using the DeepLabv3+ segmentationmodel,and identify people with GaitSet. The experimental results present that the framework achieves up to 75% accuracy of personidentification in real scenarios,which shows a significant performance improvement compared to the conventional methods. |
查看全文
查看/发表评论 下载PDF阅读器 |
关闭 |