Date of Publication :31st May 2023

Abstract: This paper discusses computer vision-based human activity recognition. The major issue is being able to identify human behavior. The main issue for video categorization systems is common human actions in videos. For instance, a running motion will be included in a long jump or running sports film. Due to its multiple applications in areas like person monitoring, human-to-object interaction, and more, human action recognition is a crucial study subject in the science of computer vision. A pre-trained CNN model for feature extraction serves as the foundation for human action recognition. Deep learning methods include convolutional neural networks (CNN). The majority of convolutional neural networks (CNNs) used for recognition tasks are constructed using convolution and pooling layers, followed by a small number of fully connected layers, and identifying similar patterns in an interval to recognize the action with accuracy of 79–90% depending on the task. The computer vision community finds the video classification problem to be very difficult. The main reason that the video categorization problem is so challenging is the shared activities that are seen in the video. A high jump sport film, for instance, combines two distinct actions—running and high jumping—that are also shown in other videos, like running or hurdling sports videos. With just one frame that captures the specific action of the event, the human brain can quickly identify the correct occurrence in a film. By removing a few significant frames from the video and using those frames to conduct the classification procedure, the same premise may also be used to video classification systems.

Reference :