Intelligent Knowledge Based Open Pose System for Video Based Activity Recognition for Alzheimer Patients

Human Activity Recognition (HAR) has become a key focus in the field of computer vision. This recognition is mainly used in the areas of driving, healthcare monitoring, sports analytics, surveillance, and human-computer interaction. The existing models are based on RGB frames or optical flow techniques. However, these methods can struggle with issues related to changes in lighting and background distractions. To address these issues, a novel approach is proposed based on the OpenPose deep learning model. This model extracts 2D human skeletal key points, which are used as essential features for predicting activities. The proposed methodology designed as an intelligent knowledge-based framework that learns patient activity. Then, the temporal modelling is implemented through neural networks. The proposed model integrates decision support system for caregivers with continuous monitoring for abnormal missed acativity. The major advantage of the OpenPose model is that it focuses on the structural dynamics of human movement instead of considering just raw pixel data. For activity classification, a Long Short-Term Memory (LSTM) is applied. It captures the sequential dynamics of these joints. Results from our experiments on standard datasets, such as UCF50 and Kinetics-400, reveal an impressive average accuracy of 96.8%, with a minimal loss of 0.38%. This significantly surpasses the performance of traditional CNN-based methods. The results underline the effectiveness of skeletal-based HAR using OpenPose in practical applications