基于FlowNet2.0改进的运动人体识别研究

Research on Improved Moving Human Body Recognition Based on FlowNet2.0

针对现有双流卷积神经网络由于运动中人体移动速度快,无法快速、准确地识别人体信息的问题,提出了一种基于FlowNet2.0网络改进的人体识别检测方法,通过给FlowNet2.0网络的各视频帧输入通道引入自注意力,能够有效增强网络对外观信息和姿态特征的提取能力,从而更好地描述运动目标。最终该模型在HDBM51数据集上进行训练,实验结果表明,改进后的FlowNet2.0网络取得了显著的改进效果。此研究为解决动作时的人体识别问题提供了一种有效的解决方案。

Aiming at the problem that the existing Two-Stream Convolutional Neural Networks cannot quickly and accurately identify human body information because the human body moves fast in motion,an improved human recognition detection method based on FlowNet2.0 network is proposed,which can effectively enhance the network's ability to extract appearance information and posture features by introducing Self-Attention into the input channels of each video frame of FlowNet2.0 network,so as to better describe moving targets.Finally,the model is trained on the HDBM51 dataset,and the experimental results show that the improved FlowNet2.0 network has achieved significant improvement results.This study provides an effective solution to solve the problems of human recognition during action.