一种基于双流卷积神经网络跌倒识别方法

针对跌倒行为的视觉特征难以提取的问题,提出一种由两路卷积神经网络和模型融合部分组成的双流卷积神经网络(Two-Stream CNN)的跌倒识别方法.该方法的一路对视频帧的运动人加框标记后,送三维卷积神经网络(3D-CNN)处理来消除视频背景的干扰;另一路从相邻视频帧获取光流图后,送VGGNet-16卷积神经网络处理;最后将3D-CNN和VGGNet-16的Softmax输出识别概率加权融合作为Two-Stream CNN输出结果.实验结果表明:标记运动人并经3D-CNN处理有效地消除了视频背景的干扰;Two-Stream CNN跌倒识别率为96%,比3D-CNN提高了4%,比VGGNet-16网络提高了3%.

A Fall Detection Method Based on Two-Stream Convolutional Neural Network

It is difficult to extract suitable visual feature for fall detection.To solve this problem,a fall detection method based on two-stream convolutional neural network(Two-Stream CNN)method was proposed.The 3-Dimensional Convolutional Neural Network(3D-CNN)stream input the marked video frame to eliminate the interference of video background.The VGGNet-16 convolutional neural network stream input the optical flow frame.Finally the Softmax of 3D-CNN and VGGNet-16 were fused as the Two-Stream CNN output.Experimental results show that,the marked video frame and 3D-CNN method can effectively eliminate the interference of the video background.The recognition rate of Two-Stream CNN is 96%,which is increased by 4% compared with 3D-CNN,3% compared with VGGNet-16 network.