Human Motion Classification Using LSTM Based on Real-World Kinect Skeletal Data

Abstract

Human motion analysis has become an important research area in physical rehabilitation applications and intelligent systems. This study aims to develop a deep neural network–based model for classifying human movements into correct and incorrect actions using skeletal data collected in a real-world environment through a Microsoft Kinect sensor. The adopted methodology involved several stages, including data collection, extraction of skeletal key points, preprocessing, and transforming the data into a relative representation to reduce the effect of differences in subjects’ positions during recording. In addition, temporal motion features were extracted. Subsequently, a model based on a Long Short-Term Memory (LSTM) network was developed to process human motion time-series sequences. The results demonstrated that the proposed model achieved good performance in motion classification, reaching an accuracy of 91.20%, along with high Precision, Recall, and F1-score values. This indicates the model’s ability to generalize and effectively handle limited real-world data. Furthermore, the study showed through comparison with the ConvLSTM model that temporally simpler models may be more suitable for small real-world datasets than more complex spatial–temporal architectures. This finding emphasizes the importance of selecting the appropriate model based on the nature of the data rather than relying solely on architectural complexity