摘要
针对传统的摄像头异常干扰识别方法识别种类单一,由预测闪烁而引起的识别准确率和可靠性低、泛化能力不强的问题,提出一种基于滚动预测平均算法的摄像头异常干扰识别方法。在自建的异常干扰图像训练集上微调ImageNet预训练的ResNet50,训练出用于摄像头异常干扰的图像分类与识别模型,在该模型的基础上运用滚动预测平均算法,以在线或离线的方式实现摄像头异常干扰视频的分类与识别。测试集实验结果表明,该方法能够正确识别出正常、遮挡、模糊和摄像头旋转视频,识别准确率达到了95%,充分验证了该方法的可行性和有效性。
Aiming at the problems that the traditional camera abnormal interference recognition methods have the characteristics of single identification type,low recognition accuracy and reliability,and poor generalization ability due to the prediction of flic-ker,a method for camera abnormal interference recognition based on the rolling prediction average algorithm was proposed.The ImageNet pre-trained ResNet50 on the self-built abnormal interference image training set was fine-tuned.A model for camera abnormal interference image classification and recognition was trained.The rolling prediction average algorithm was applied to the model to realize the classification and recognition of the camera abnormal interference video online or offline.Experimental results on the test set show that the proposed method can correctly identify the normal videos,occlusion videos,blurred videos and the camera rotated videos.The recognition accuracy reaches 95%,which fully verifies the feasibility and effectiveness of the proposed method.
作者
杨亚虎
陈天华
邢素霞
王瑜
YANG Ya-hu;CHEN Tian-hua;XING Su-xia;WANG Yu(School of Computer and Information Engineering,Beijing Technology and Business University,Beijing 100048,China)
出处
《计算机工程与设计》
北大核心
2021年第1期248-255,共8页
Computer Engineering and Design
基金
国家自然科学基金项目(61671028)。
关键词
智能视频监控
卷积神经网络
深度学习
智慧城市
公共安全
intelligent video surveillance
convolutional neural network
deep learning
smart city
public safety