This research presents applications of novel traffic data, which could be collected through connected-automated vehicles(CAVs), in traffic management of urban freeways. To this end, firstly, this research selects bottleneck sections on Seoul urban freeways and performs a Survival Analysis for each bottleneck to derive the relationship between the flow rate and probability of traffic congestion. The analysis results reveal that the freeway congestion in Seoul occurs at a lower flow rate than expected, and each bottleneck has its unique characteristics in terms of congestion probability. These findings can provide the statistical features of freeway bottlenecks. However, it shows a huge limitation of uncertainty to be applied in traffic management in the real world since traffic congestion can also occur due to other reasons, such as unusual driver behaviours.

To overcome this issue and determine the potential of new traffic data from CAVs, this research collects novel traffic data based on vehicle trajectories. Specifically, we use drones to record vehicle behaviours near bottlenecks when congestion occurs and apply artificial intelligence technologies to extract vehicle trajectories every 0.1 seconds. Based on the data from drone video analysis, we can estimate the traffic state over each time-space area of freeways using Edie’s Generalized Definition, and construct a fundamental diagram(FD) that presents specific characteristics of the traffic flow. By tracking the traffic state changes on the FD according to the occurrence of congestion, we can classify the process of congestion for the three phases as follows: (i) increasing the flow rate in a free-flow state, (ii) local congestion and recovery, and (iii) overall congestion and propagation. We also gain several insights into traffic management as the importance of pro-active traffic management for maintaining a higher bottleneck capacity, and the necessity of post-active strategies for preventing aggravation of congestion.

To clearly present the degree of congestion, which occurs at the extensive time-space area, with vehicle trajectory data, this research also develops novel congestion indicators as ‘Residual’ in terms of space and ‘Delay’ in terms of time. Specifically, we define the ‘Residual’ and ‘Delay’ as the difference between the actual and expected trajectories collected and estimated from drones and the latest car-following models, respectively. These new indicators can be applied at both the vehicle-level and road-level to describe the degree of congestion by the accumulation value over time. From the distribution of the driver reaction time derived during the development of the indicator, we can also identify the characteristics of Seoul drivers and estimate the difference between the ideal and actual road capacity.