【摘要】 基于光流法和卷积神经网络,提出一种室内跌倒行为检测方法。在数据预处理方面,使用光流法把监控视频转化为由光流图像组成的动作序列;在模型方面,采用卷积神经网络VGG-16对输入动作序列进行训练和优化,根据softmax分类器输出结果不断调整权重和偏量;在训练过程中,制作了跌倒数据集并采用迁移学习训练策略解决训练过程中跌倒行为数据过少的问题。实验结果表明该检测方法可在室内环境下正确地检测出跌倒行为的发生,达到了较高的准确率。更多还原

【Abstract】 A method of indoor falling behavior detection based on optical flow and convolution neural network is proposed. In the aspect of data preprocessing, the optical flow is used to transform the surveillance video into an action sequence composed of optical flow images. In the aspect of model, the input action sequence is trained and optimized by using convolutional neural network VGG-16, and the weights and offsets are constantly adjusted according to the output results of softmax classifier. In the training process, the falling data set is made and the transfer learning training strategy is adopted to solve the problem of too little falling behavior data in the training process. The experimental results show that the detection method can correctly detect the occurrence of falling behavior in indoor environment and achieve high accuracy.更多还原