Elderly Exercise Activity Classification Based on Pose Estimation Using CNN-LSTM

Elderly people who exercise require specialized programs designed for them, as well as help from physiotherapists or other professionals. However, for a variety of reasons, older people frequently do not receive support from physiotherapists. One of them is that elderly people have more difficulty accessing exercise because there are fewer physiotherapists than there are elderly people. This research proposed a deep learning-based approach to classifying elderly exercise activities. We proposed a method that uses the Mediapipe Pose Estimation (MPE) framework to estimate the pose of the elderly. Then, the sequences of exercise activities are trained using the Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). This method is used for labeling and classifying the exercise types performed. In addition, we introduced an exercise activity dataset for the elderly since simple exercise activities are not commonly found. The dataset used is a collection of physiotherapy exercise videos for the elderly, based on preselected classes of exercises. The output of this model is a model that can classify 9 classes of elderly exercise activities. The accuracy of the resulting model is 96.30%.