Urban road traffic state recognition based on traffic video

Accurate traffic state recognition is essential for the development of intelligent transportation systems (ITS). However, existing approaches often rely on limited traffic indicators and may not fully exploit the spatiotemporal information available in surveillance video data. This paper proposes a real-time traffic state recognition model based on a Bayesian-Optimized Multi-Kernel Multi-class Support Vector Machine (BO-MK-MSVM), leveraging urban surveillance videos. The method integrates YOLOv8 for vehicle detection with DeepSORT for multi-object tracking, enabling the extraction of key traffic parameters—flow, speed, and density—at two-minute intervals. Fuzzy C-Means (FCM) clustering is employed to generate labeled data for four traffic states. A multi-kernel support vector machine (MK-MSVM), combining radial basis function and polynomial kernels, is used to model the nonlinear distribution of traffic features. Bayesian optimization is applied to tune the model’s hyperparameters, resulting in the final BO-MK-MSVM classifier. Experiments conducted on both the UA-DETRAC dataset and a real-world dataset from Lanzhou demonstrate that the proposed model achieves high accuracy (99.29%) and F1-score (99.21%) while supporting real-time inference. Additionally, tests under varying illumination conditions confirm the robustness of the model in different lighting environments. The results indicate that BO-MK-MSVM outperforms conventional single-kernel SVM and CNN-LSTM models, offering a robust and scalable solution for urban traffic state monitoring.