摘要
异常行为检测是智能安全监控的重要内容,而异常行为特征的提取是一个难点.张量作为高维数据的自然表现形式能有效保留数据的结构信息提取到运动目标.本文将张量分解应用于数字视频处理,然后对稀疏前景张量时间轴方向上的纤维束做频域处理进一步优化运动前景;最后使用多层卷积神经网络结构对运动目标的跑动行为进行识别.仿真对比实验证明了基于张量分解的方法比传统方法处理效果更好、在实测视频中本文跑动行为识别率达到81.4%.
Anomaly detection is an important part of intelligent security monitoring,and the extraction of abnormal behavior characteristics is relatively difficult.As a natural representation of high-dimensional data,tensors can effectively preserve the structural information of data and extract the moving objects in video.In this paper,tensor decomposition is applied to digital video processing,and then the frequency domain of sparse tensor is processed in a frequency domain to further optimize the motion foreground.Finally,a multilayer convolutional neural network structure is used to train the classifier to detect the running behavior of moving targets.The simulation experiment shows that the tensor decomposition method is better than the traditional method,and the detection rate of running behavior is 81.4% in the realistic scene.
出处
《五邑大学学报（自然科学版）》
CAS
2017年第4期49-56,共8页
Journal of Wuyi University(Natural Science Edition)
关键词
张量分解
高维信号处理
卷积神经网络
跑动检测
tensor decomposition
high dimensional signal processing
convolution neural networks
running detection
