Human flow detection using CSI technology combined with STFT and machine learning

This study investigates the significance of pedestrian flow detection technologies in highly populated areas such as museums, exhibition centers, and amusement parks, particularly in the fields of enterprise management, smart healthcare, and public safety. Traditional detection methods, such as cameras and infrared sensors, are often constrained by privacy concerns and environmental factors. In contrast, Channel State Information (CSI) technology utilizes variations in wireless communication signals, offering a privacy-preserving and cost-effective solution. To validate this approach, the study employs lightweight WiFi transceivers to capture signal perturbations caused by human activities, with a custom labeling and balancing method applied to the collected data. Short-Time Fourier Transform (STFT) is then used to convert the data into timefrequency domain representations for feature extraction. The processed dataset is subsequently fed into machine learning models for training and prediction. Four machine learning algorithms—Random Forest Classifier (RandomForestClassifier), Support Vector Classifier (SVC), XGBoost Classifier (XGBClassifier), and Gradient Boosting Classifier (GradientBoostingClassifier)—were evaluated, all demonstrating excellent performance. Among these, the Random Forest Classifier achieved 99% accuracy in scenarios detecting 0–2 people passing through the monitored area. The results indicate that integrating WiFi-based CSI technology with machine learning models can enable precise and efficient real-time pedestrian flow monitoring, showcasing promising applications in museum and healthcare environments.