A Deep Learning-Based Framework for Urban Traffic Monitoring and Short-Term Flow Prediction Using YOLOv12 and CNN-Bi-LSTM

Rapid urbanization and the continuous growth in the number of vehicles have made traffic congestion a significant challenge for modern transportation systems. Reliable traffic monitoring and short-term traffic-flow prediction can help in understanding changing road conditions and support more effective traffic management. This paper presents a deep learning-based framework for urban traffic monitoring and short-term traffic-flow prediction using YOLOv12 and CNN-Bi-LSTM. In the proposed system, YOLOv12 processes traffic video frames to detect and classify vehicles and provides the information required for extracting traffic-related features such as vehicle count, vehicle type, estimated speed, traffic density, and traffic flow. The resulting observations are organized chronologically to form a traffic time-series dataset. These sequential features are then provided to a CNN-Bi-LSTM model, where the convolutional component learns relevant local patterns and the Bi-LSTM component captures temporal relationships within the traffic sequence. The model generates short-term traffic-flow predictions based on previously observed traffic conditions. The prediction performance is evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE), while vehicle-detection performance is assessed using appropriate detection measures such as precision, recall, F1-score, and mAP. The proposed framework combines computer-vision-based traffic monitoring with temporal deep learning to provide an integrated approach for analyzing and forecasting urban traffic conditions.