A Survey and High-Level Design on Human Activity Recognition

Human Activity Recognition is one of the active research areas in Computer Science. The domain is being recognized as it has a vast scope and immense potential. The resultant applications are being deployed in various contexts like security surveillance, healthcare, human-computer interaction, Traffic Control, and many more, will help minimize human errors and associated costs and increase the efficiency of the desired work. This work surveyed recent eleven research papers and the relevant topics in the Human Activity Recognition (HAR) domain. Different applications, such as detection of single or multiple persons, classification of simple activities, anomalous activities, tracking, etc., with proposed related technologies, are covered. Based on the review and the conclusion got, a feasible vision-based model to detect, classify and recognize human activities or actions has been proposed. Techniques like Haar feature-based technique [10] [11] for the detection of human beings and feature extraction, YOLO [5], CNN (Resnet) [1] [2] [4] The classification and recognition of the defined activities have been proposed to be used. The design includes sub-systems such as human detection and feature extraction, pre-processing of the frames, classification of actions using Deep Neural Network, and raising warnings. The system has been proposed to work within the limited computational capabilities to determine anomalous activities.