摘要
提出了一种优化预测网络的视频异常行为检测方法,现有的大部分视频异常行为检测方法主要基于最小化重建误差准则,缺乏对异常行为运动长期性的考虑。因此,提出了一种基于运动上下文感知的预测网络来进行视频异常行为检测。上述方法引入具有记忆对齐学习的长期运动上下文记忆,能够将视频帧的信息存储到存储器中,建立起当前输入帧与前后帧的映射关系以方便后续的预测,通过预测未来帧与原始帧之间的差异来检测视频目标异常行为。同时,采用具有重叠注意力的全局匹配光流估计方法来获取未来帧和原始帧的光流图,并在预测中引入运动约束让未来帧和原始帧之间的光流保持一致。此外,采用生成性对抗网络进行训练模型,让上述方法能更好地预测未来帧,提高视频异常行为检测的准确性。Avenue和ShanghaiTech两个数据集上的实验结果表明,所提方法相比于其它的视频人体异常行为检测方法有更好地表现,在Avenue和ShanghaiTech数据集中分别达到了90.7%和85.4%。
A video abnormal behavior detection method based on optimized prediction network is proposed.Most of the existing video abnormal behavior detection methods are mainly based on the criterion of minimizing reconstruction error,and lack of consideration of the long-term movement of abnormal behavior.Therefore,a prediction network based on motion context awareness is proposed to detect abnormal video behavior.This method introduces long-term motion context memory with memory alignment learning,which can store the information of video frames in memory,establish the mapping relationship between the current input frame and the previous and subsequent frames to facilitate subsequent prediction,and detect the abnormal behavior of video targets by predicting the difference between future frames and original frames.At the same time,the global matching optical flow estimation method with overlapping attention is used to obtain the optical flow diagram of the future frame and the original frame,and the motion constraint is introduced in the prediction to keep the optical flow between the future frame and the original frame consistent.In addition,the training model is based on the generative confrontation network,which makes this method better predict future frames and improve the accuracy of video abnormal behavior detection.The experimental results on Avenue and ShanghaiTech data sets show that this method performs better than other video human abnormal behavior detection methods,reaching 90.7%and 85.4%in Avenue and ShanghaiTech data sets respectively.
作者
庄旭
张红民
郑敬添
ZHAUNG Xu;ZHANG Hong-min;ZHENG Jing-tian(School of Electrical and Electronic Engineering,Chongqing University of Technology,Chongqing 400054,China)
出处
《计算机仿真》
2025年第5期475-481,共7页
Computer Simulation
基金
国家自然科学基金资助项目(61901068)
重庆市自然科学基金面上项目资助(cstc2021jcyj-msxmX0525)
重庆市自然科学基金面上项目资助(CSTB2022NSCQ-MSX0786)。
关键词
视频异常行为检测
运动上下文感知
长期运动上下文记忆
光流估计
生成性对抗网络
Video abnormal behavior detection
Motion context awareness
Long-term motor context memory
Optical flow estimation
Generative adversarial network
作者简介
庄旭(1999-),男(汉族),湖北省孝感市人,硕士研究生,主要研究领域为信号与信息处理;张红民(1970-),男(汉族),河南省舞阳人,博士,教授,主要从事图像处理与模式识别方面的研究;郑敬添(1998-),男(汉族),山东省济南人,硕士研究生,主要研究领域为信号与信息处理。
