Research Of Action Recognition From Videos Using Deep Neural Networks

Action recognition is a popular research field in recent years.Being able to recognize the action in the video will have a profound impact on society.It can play a huge role in intelligent monitoring,video retrieval,medical health,human-computer interaction,etc.Action recognition is also subject to various restrictions,which affects the accuracy of behavior recognition.It cannot completely rely on action recognition technology to solve the problems in life.In order to improve the accuracy of action recognition,a model based on the combination of residual network and long-short-term memory network is designed and implemented.The video contains spatial features and temporal features.These two pieces of information are about the performance of behaviors.These two features are calculated using deep learning,and the extracted information is fused to obtain the classification result of the video.First,the video is decomposed into image frames,and a part of the frames are extracted as the input of the residual network.The video frame contains spatial features.The residual network with excellent performance in the image data set is used to extract the spatial features in the video.The process can be seen as the encoding of video frames;after the spatial features are extracted,the spatial features are input into the long and short-term memory network,and the network's ability to deal with dependent problems is used to extract the temporal information in the video,and the temporal features are extracted from the video It can be regarded as a decoding process,and the output of the temporal network is fully connected at the end to obtain the classification result of the video behavior of the model.In the UCF-101 dataset,the accuracy rate of the model for action recognition reached 85.68%,and the baseline accuracy rate of action recognition was 43.9%.The experimental results want to improve the accuracy rate of other classic models.The experimental model can better complete the task of behavior recognition in the data set.