Human activity recognition using hybrid model
2022, vol.14 , no.4, pp. 55-66
The main task of the Human Activity Recognition (HAR) is to recognize the different actions performed by the individual. There are various machine learning and deep learning models which have been presented to predict the given activity of the human. The problem of the existing system is that they could not identify the activity when there is a sudden change in the activity. Hence, this paper proposes a model by using the deep learning concepts which can predict the activities and also can predict the sudden transition of one activity to another. This method uses the Convolution-Neural-Network Layer (CNN) and Gated-Recurrent-Unit (GRU) which identifies the changes in the activities collected by the sensors and gives us correct results. For the experimentation of this model, the University-of-California (UCI) Human Activity Recognition dataset has been used. This dataset comprises of various activities such as walking, sitting, standing etc. The results have compared with Human Activity Recognition System (HARS). The proposed model attained an accuracy of 96.79% whereas the HARS attained 95.99%. When compared with precision, recall and F1-score, the proposed model performed better than the existing model.
Monitoring, HAR, CNN, GRU, UCI
Srikanta Swamy Pushpalatha, Shrishail Math. Human activity recognition using hybrid model. International Journal on Information Technologies and Security, vol.14 , no.4, 2022, pp. 55-66.