摘要
为解决安防领域目标检测中复杂背景信息对检测系统的干扰并满足检测系统实时性的要求,提出一种基于深度学习在复杂背景下的移动目标检测算法。为减小复杂背景信息对检测系统的干扰,采用混合高斯背景建模算法(GMM)将视频图像中的前景与背景信息分离,提取出有效的前景位置信息。为提升算法模型的检测速度,采集数据集训练YOLOv4算法,接着将训练后的算法模型输入到剪枝网络进行模型剪枝,得到模型参数少、检测速度快的改进YOLOv4算法模型。最后,将GMM算法提取的前景位置信息作为改进的YOLOv4算法的检测输入,实现对复杂背景下移动目标的快速检测。根据实验结果所示,所提出的算法在测试数据上背景信息造成的误检降低了71.7%,检测时间减少了32.9%。实验结果表明,所提出的算法能够有效减少复杂背景信息对检测系统的干扰,同时算法具有较高的检测速度。
In order to avoid the interference of complex background information on the detection system and meet the requirements of real-time performance of the detection system in target detection in the field of security and protection,a moving target detection algorithm based on deep learning in complex background is proposed. In order to reduce the interference of complex background information on the detection system,the Gaussian mixture model(GMM)algorithm is used to separate the foreground information from background information in the video,so as to extract the effective foreground position information. In order to increase the detection speed of the algorithm model,a data set is collected to train the YOLOv4 algorithm,and then the trained algorithm model is input into the pruning network for model pruning. Finally,the improved YOLOv4 algorithm model with fewer model parameters and faster detection speed is obtained. The foreground position information extracted by the GMM algorithm is used as the detection input of the improved YOLOv4 algorithm to achieve rapid detection of moving targets in complex backgrounds. According to experiments,the proposed algorithm′s false detection caused by the background information in the test data is reduced by 71.7% and its detection time is shortened by 32.9%. The experimental results show that the proposed algorithm can effectively reduce the interference of complex background information on the detection system. In addition,the algorithm has higher detection speed.
作者
邬军
邓月明
何鑫
李小军
石韧
WU Jun;DENG Yueming;HE Xin;LI Xiaojun;SHI Ren(College of Information Science and Engineering,Hunan Normal University,Changsha 410081,China;Hunan Novasky Electronic Technology Co.,Ltd.,Changsha 410221,China)
出处
《现代电子技术》
2022年第19期59-65,共7页
Modern Electronics Technique
基金
国家自然科学基金项目(62173140)
国家自然科学基金项目(62072175)
湖南省重点研发计划项目(2022GK2067)
湖南省自然科学基金项目(2021JJ30452)
湖南省教育厅科学研究项目(21C0008)。
作者简介
邬军(1994—),男,河南信阳人,硕士研究生,主要研究方向为目标检测;通讯作者:邓月明(1981—),男,湖南邵阳人,高级实验师,硕士生导师,主要研究方向为智能系统与边缘计算;何鑫(1987—),男,湖南邵阳人,博士,主要研究方向为光电工程和深度学习;李小军(1998—),男,湖南湘潭人,硕士研究生,主要研究方向为目标检测;石韧(2000—),男,湖南邵阳人,主要研究方向为智能信号处理。
