Activity Recognition in Video Frames for Enhancing Human-Robot Collaboration: A Machine Learning Perspective

Abstract

Human activity recognition is a prominent area of investigation within computer vision research. Its potential applications span diverse fields such as security monitoring, healthcare, and human-computer interaction. The identification and classification of human activity through the utilization of machine learning techniques has yielded a plethora of proposed models within this dynamic field of scientific inquiry. By employing diverse kinetic models that are linked to the learning of spatial or temporal features, the characteristics of a collection of image or video data are extracted. The primary aim of this model is to identify and classify ongoing activities. This objective has been successfully achieved by various deep-layer trained models that have been implemented in this field. To fulfill the specified criteria, the proposed approach utilizes the Long Recurrent Convolutional Network (LRCN) and Long Short-Term Memory (LSTM) convolutional neural network architectures. The implementation of Vision-Based Human Activity Recognition utilizes the 3D Convolutional Neural Network (CNN) model. The model undergoes training utilizing the UCF-50 dataset. The classification of the photos by the model is contingent upon the action being performed. The model will receive videos as input, which will undergo appropriate pre-processing before being classified based on the activity. The vision-based human activity identification system offers several advantages, including but not limited to accuracy, cost-effectiveness, and user-friendliness.