基于卷积神经网络的人群突散异常行为检测

Detection of abrupt dispersal of abnormal human behavior based on convolutional neural network

为检测人群突散异常,提出一种基于卷积神经网络的人群突散异常行为检测方法。对于人群中的个体使用改进的多尺度卷积神经网络(MCNN)预测人群中每一个个体头部的坐标位置;根据提取出来的坐标点计算人群平均动能、人群密度值以及人群分布熵这3种人群运动状态特征值,以此减少计算量;将3种运动状态特征值放入基于差分进化粒子群优化的极限学习机(DE-PSO-ELM)中进行训练预测,得到人群运动状态,实现人群突散异常行为的检测。仿真结果表明,该算法对人群突散异常行为检测有较好的效果,检测准确率达到99.75%。

To detect population spurting anomalies,a method for population spurting abnormal behavior detection based on con-volutional neural network was proposed.An improved multi-scale convolutional neural network(MCNN)was used to predict the coordinate position of each individual head in the crowd.The average kinetic energy,the density value,and the distribution entropy of the crowd were calculated according to the extracted coordinate points.The eigenvalues of the three motion states were put into the extreme learning machine based on differential evolution particle swarm optimization(DE-PSO-ELM)for trai-ning and predicting,so as to obtain the motion state of the crowd,and the detection of spurting behavior abnormality of the crowd was realized.The simulation experiment results show that the proposed algorithm has good effects on the detection of spurting behavior abnormality of the crowd,and the detection accuracy rate reaches 99.75%.