人体运动视频图像目标局部特征快速提取仿真

Fast Extraction And Simulation of Local Features of Human Body Moving Video Image

针对当前人体运动视频图像目标局部特征快速提取时,视频图像目标局部参数分析存在不足,导致构建视频图像阶层时对低级层特征图描述不精准,从而增加提取时间,存在查全率和查准率较低等问题。提出人体运动视频图像目标局部特征快速提取方法。通过人体运动视频图像相邻帧之间的信息熵,结合密度函数确定视频图像初始聚类中心,计算得到聚类数目,实现人体运动视频图像目标局部参数分析。选取卷积输出层的特征图构建人体运动视频图像阶层结构,引入信息熵的方法对人体运动视频图像低级层特征图进行描述,结合区域平均的方法描述人体运动视频图像高级层特征图,最终完成人体运动视频图像目标局部特征提取。实验结果表明,所提出方法在人体运动视频图像目标局部快速提取过程中,特征提取所需完成时间较短,查全率和查准率较高。

When the local features of human moving video images are extracted quickly,there are some shortcomings in the analysis of local parameters of video images,which leads to the inaccurate description of low-level feature maps in the construction of video image strata,thus increasing the extraction time and low recall and precision.An image local feature extraction method based on convolutional neural network is proposed.Through the information entropy between adjacent frames of the moving video image of the human body,the initial clustering center of the video image is determined by combining the density function.The number of clusters is calculated,and the analysis of local parameters of human motion video image is realized based on the local characteristic parameters of the human motion video image.According to the characteristic parameters,a video image feature map of human motion is generated,and the feature map of the convolution output layer is used to construct the hierarchical structure of the human motion video image.The method of information entropy is introduced to describe the lower level feature map of human motion video image,and the method of region average is introduced to describe the high level and feature map of human motion video image.Finally,the local feature extraction of human motion video image with strong expression ability is constructed.The simulation results show that the proposed method has shorter completion time for feature extraction and higher recall rate and precision in the process of local extraction of human motion video images.