Human flow recognition is crucial for applications such as crowd management, safety monitoring and behaviour analysis, but traditional visual-based methods face challenges like occlusion, environmental interference and privacy concerns. This study proposes a novel approach utilising physiological signals captured from Emotiv, Empatica and RespiBAN wearable devices. The research investigates human flow pattern recognition through multimodal signals, including heart activity, electrodermal activity and respiration, classified using Long Short-Term Memory (LSTM) deep learning models. These LSTM models are designed to capture long-term dependencies in temporal data. The results show that the LSTM-based approach outperforms conventional machine learning models, such as SVM and Decision Trees, achieving a classification accuracy of 90.7%. Thus, the obtained results demonstrate the potential of physiological signals as non-invasive, robust indicators of human movement, providing a reliable alternative when traditional vision-based methods are not feasible.