In response to the problem that the analysis of highway congestion in hourly units fails to accurately reflect real-time service levels, this study investigates a method for real-time grading of traffic congestion levels on highways based on video surveillance. The research employs YOLOv5 and the Deep SORT algorithm to extract vehicle trajectories and compute short-term traffic flow macro parameters (including flow rate, density, and speed) and micro parameters such as headway intervals on a one-minute basis. Using the entropy method, weights are assigned to indicators that characterize congestion states to construct a comprehensive congestion measurement index. This index is then used for real-time congestion level classification through Fuzzy C-Means (FCM) clustering, with validation performed using K-means algorithm. The results indicate that the proposed comprehensive congestion measurement index outperforms congestion indicators constructed solely from macro and micro parameters, providing a real-time assessment of traffic flow congestion status for different time periods.