The objective of this project was to demonstrate how vehicle trajectory data can be obtained from an existing radar-based vehicle detection system and used to produce turning movement count reports at signalized intersections with both exclusive and shared lanes. Vehicle trajectory data and video recordings were obtained at a main location in Appleton, Wisconsin, and at two supplemental locations (Appleton and Madison, Wisconsin). The supplemental locations were used to better understand how the data collection and algorithms developed perform under varying geometric conditions. An algorithm that processes vehicle trajectory data collected from a radar device and generates turning movement counts was developed and implemented in the R programming language. The algorithm relies on vehicle trajectories downstream of an automatically detected stop bar to classify vehicle movements into left, thru, and right movements. The actual number of vehicles was obtained by performing manual turning movement counts using intersection video. The stop bar position plays a key role in removing noise in the dataset, such as vehicles in nearby parking lots. When the number of vehicles detected by the algorithm is compared with the number of vehicles from a manual count the results indicate an average accuracy of over 99%. A more detailed analysis suggests that the average difference between the number of vehicles classified as making a specific movement during a 15-minute period and the actual number of vehicles in the same period is ± 2 in more than 60% of the periods evaluated. The evaluation of the algorithm performance in 15-minute intervals, regardless of traffic volumes, provides a more intellectually honest evaluation of the results by moving away from the standard practice of reporting vehicle detection system performance using large volumes and ignoring turning movement breakdown. Coincidentally, when the performance of the developed algorithm is evaluated under volume conditions that approach 100 vehicles per movement during a 15-minute period, the results approach accuracy levels greater than 90%. Since the algorithm relies on data from a vehicle detection system, the performance can degrade (as was found in the supplemental data collection sites) when the line of sight between vehicles and the detection system is interrupted (vehicle not visible). To successfully commercialize this innovation, the algorithm needs to be improved and a prototype data collection device that can be installed inside a signal cabinet and a centralized software tool to manage multiple data collection devices need to be developed. Although the results from the project are encouraging, algorithm improvements are needed in order to have a market-ready solution. Modifications will require data collection across different locations in the country in order to obtain vehicle trajectory datasets from a wide array of geometric and traffic conditions. These modifications will in turn require further analysis and performance evaluations as part of an iterative development process. In terms of data collection, the system should be tested on a smaller hardware platform such as a development board and configured for an environment in which the data analysis algorithm and data collection system are on separate locations. All these improvements and modifications are planned in a follow-on NCHRP IDEA project.
The final report is available.