چكيده به لاتين
For metro traffic modeling and control, continuous monitoring of congestion levels and passenger counting from station images is essential. The primary goal of this study is to develop advanced algorithms such as YOLOv8 for accurate passenger counting in metro stations. Initially, YOLOv7 was employed to identify and track passengers in station images. Although YOLOv7 demonstrated acceptable accuracy, it faced challenges such as misidentifying human-like signs as passengers, failing to distinguish metro staff from passengers, and counting passengers inside trains as those waiting at the station.
To address these challenges, a new model was designed and trained using the YOLOv8 algorithm, which offers higher accuracy and efficiency. The required training images were collected and pre-processed, including format conversion, normalization, feature extraction, cropping, and noise removal. These images were labeled into categories such as passengers, staff, and trains using the Roboflow platform, and the YOLOv8 model was trained accordingly. YOLOv8 exhibited superior speed and accuracy in identifying passengers and staff compared to YOLOv7.
Furthermore, the k-means clustering algorithm was utilized to classify passengers based on congestion levels into categories such as very crowded, crowded, moderate, sparse, and very sparse. The trained software, leveraging the YOLOv8 region-based model, identified trains and subsequently distinguished, tracked, and counted passengers and staff separately. Upon completing the counting process, the number of individuals counted in every 150 frames (5 seconds) in real time was fed into a traffic model for a metro line.
The results demonstrated that YOLOv8 significantly outperformed YOLOv7 in accurately identifying, tracking, and counting passengers and staff. Additionally, analysis of the charts and parameters derived from the traffic model confirmed the success of YOLOv8 in providing the required data (real-time passenger numbers), leading to improved accuracy, speed, and ultimately the advancement of the traffic modeling system.