Gait Based Human Activity Recognition using Hybrid Neural Networks

Human activity recognition (HAR) has been used in a range of fields, including security and surveillance, human-robot interaction, and entertainment, thanks to the quick growth of IoT and DL technology. In this research paper, the models CNN, LSTM, GRU, CNN-LSTM, and CNN-GRU are evaluated, and the most accurate model is determined for human activity recognition. The sensor values are initially gathered for each individual while they are engaged in a variety of activities. On the UCI HAR public dataset, the CNN-GRU model achieves state-of-the-art performance of 91% accuracy. Its performance is shown to be superior to other deep neural network architectures and machine learning models that have been suggested in the past by comparing its outcomes to those of other methodologies. The suggested method is reliable and has a variety of uses, including in the fields of security, sports, and healthcare. According to the findings, the hybrid CNN-GRU model that has been proposed may offer a practical method for identifying human activity from gait analysis.