基于卷积神经网络嵌套模型的人群异常行为检测

ABNORMAL BEHAVIOR DETECTION OF CROWDS BASED ON NESTED MODEL OF CONVOLUTIONAL NEURAL NETWORK

为了提高对运动目标的精确提取,减少冗余特征信息,提升算法的泛化性能和非线性拟合能力,提出基于卷积神经网络嵌套模型的人群异常行为检测方法。通过嵌套mlpconv层改进卷积神经网络结构,利用混合高斯模型有效、精确地提取出视频中前景目标。嵌套多层的mlpconv层自动学习前景目标的深度层次特征,生成的特征图经过向量化处理输入到与全连接层相连的Softmax分类器进行人群中异常行为检测。仿真实验结果表明,该算法减少了对冗余信息的获取,缩短了算法运算时间和学习时间,改进的卷积神经网络在泛化性能和非线性拟合能力都有提高,对人群异常行为检测取得较高准确率。

In order to improve the accurate extraction of moving objects, reduce redundant feature information extraction, improve the generalization performance and non-linear fitting ability of the algorithm, we proposed an abnormal behavior detection of crowds based on nested model of convolutional neural network. This method improved the structure of convolution neural network by nesting mlpconv layer, extracted foreground objects in video effectively and accurately by using mixed Gaussian model, and nested multi-layer mlpconv layer to automatically learn the depth features of foreground objects. The generated feature map was vectorized and input into the Softmax classifier connected with the full connection layer to detect abnormal behavior in the crowd. The simulation experimental results show that the algorithm reduces the acquisition of redundant information, shortens the computation time and learning time of the algorithm, improves the generalization performance and non-linear fitting ability of the improved convolutional neural network, and achieves a high accuracy for crowd abnormal behavior detection.