基于YOLOv5的边境可疑人员检测系统设计
Design of Border Suspicious Person Detection System Based on YOLOv5
岳廷树 1鄢元霞 1潘文林2
作者信息
- 1. 云南民族大学电气信息工程学院 昆明 650504
- 2. 云南民族大学数学与计算机科学学院 昆明 650504
- 折叠
摘要
针对传统边境监控系统需要依靠大量边防人员,且人力在可见光检测中存在识别不明的情况,提出基于YO-LOv5s目标检测算法的Android红外检测系统,该系统可在无网络情况下运行.以无人机拍摄的红外数据图像为样本,模拟边境不同场景下的可疑人员入侵,并利用深度学习算法进行训练,不同场景下的检测准确率均可达到95%左右,通过ONNX模型交换格式转换成易于终端部署的NCNN模型,并利用Android Studio进行Android端模型部署,最终实现手机端的红外图像检测.实验结果表明,部署于手机端的红外检测模型对边境周边可疑人员的单帧图像检测时间低于200 ms,检测置信度达到94%左右,基本符合系统预期目标.
Abstract
Aiming at the fact that the traditional border monitoring system needs to rely on a large number of border guards,and the manpower is unclear in the visible light detection,an Android infrared detection system based on the YOLOv5s target detec-tion algorithm is proposed,which can run without a network.Taking the infrared data images captured by drones as samples,simu-lating the intrusion of suspicious persons in different scenarios at the border,and using deep learning algorithms for training,the de-tection accuracy in different scenarios can reach about 95%.Through ONNX model exchange format conversion,the NCNN model is easy to be deployed on the terminal,and Android Studio is used to deploy the model on the Android side,and finally the infrared im-age detection on the mobile phone side is realized.The experimental results show that the detection time of the infrared detection model deployed on the mobile phone for a single frame of suspicious persons around the border is less than 200 ms,and the detec-tion confidence reaches about 94%,which basically meets the expected goals of the system.
关键词
深度学习/YOLOv5s/红外检测/模型转换/AndroidKey words
deep learning/YOLOv5s/IR detection/model conversion/Android引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据