| 张振亚,屈兴亮,王萍,方博,何明艳,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(): |
| 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 Target Identification Method in Building Based on Gait Recognition |
| 投稿时间：2022-08-28 修订日期：2022-10-20 |
| 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重点项目(KJ2020A0470)；2022年高校优秀青年人才支持计划项目(gxyq2022030)；安徽省高校学科拔尖人才学术资助项目(gxbjZD2021067)；安徽省特支计划创新领军人才项目。 |
|
| 摘要点击次数: 1758 |
| 全文下载次数: 587 |
| 中文摘要: |
| 人员的身份辨识及其时空分布是建筑智能化中经常需要的一类基础输入信息。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到的建筑物内的视频流,该框架首先利用yolov5和DeepSort实现了对视频中出现的多个人员步态序列的构建,然后利用DeepLabv3+分割模型实现了人员步态序列图像的分割,再利用GaitSet实现了人员身份的辨识。实验表明,所设计的框架方法在真实场景下人员辨识准确率可以达到75%,与常用方法相比,性能提升显著。 |
| 英文摘要: |
| The identification of occupants and their spatial and temporal distribution represents a basic input frequently required in building intellectualization. Therefore, in order to effectively resolve the problem of simultaneous identification of multiple occupants in buildings, this paper proposed a multi-occupant identification that has been designed based on GaitSet (MOIG) framework for gait recognition in buildings. Regarding the video stream inside the building captured by the camera, the framework first utilizes yolov5 and DeepSort to construct multiple gait sequences of people appearing in the video, then it implements the segmentation of the gait sequence images by using the DeepLabv3+ segmentation model. Subsequently, GaitSet is utilized for people identification. The experimental results present that the designed framework method will be able to successfully achieve 75% accuracy in real scenes, and its performance has been significantly enhanced as compared with the common methods. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
