Research and Analysis of Action Recognition Based on Video

Abstract

In recent years, along with the swift development of computer vision and deep learning technologies, videobased action recognition has turned into one of the core research directions within the field of artificial intelligence. Its accomplishments are extensively applied in such real scenarios as intelligent monitoring, human-computer interaction, and autonomous driving. This paper first presents the research background and practical significance of video-based action recognition in recent years then analyzes the current challenges such as those of complex background motion blur and similarity among categories. Next it expounds upon the principle’s structures and application effects of representative models such as 3D convolutional neural networks two - stream networks and models based on the Transformer and also analyzes the advantages and disadvantages of various models. Finally, it summarizes the application scenarios of video action recognition, explores the existing technical difficulties, and also looks forward to the development trends in the future like lightweight models and few - shot learning. This paper offers comprehensive references for researchers in relevant fields, making them able to grasp the research current situation and carry out in - depth research.