A comparison of machine learning-based method vs. the Highway Capacity Manual method of intersection delay

This paper offers a comparative study of traffic delay times at intersections using two methodologies: the established Highway Capacity Manual (HCM) methodology and a machine learning vision system (MLVS). Two intersections in Tuscaloosa, Alabama—Hackberry Lane and Peter Bryce Boulevard, known for substantial traffic volume and proximity to the University of Alabama, and the newly established 2nd Avenue and Coliseum Circle—served as the empirical settings of the study. The goal of this paper was to examine the feasibility of automated delay measurement and level of service reporting. For the second intersection, paired t-test was utilized to evaluate the variations in delay times produced by the HCM method and MLVS. The analysis used was able to identify statistically significant similarities and differences at the intersections between the two methodologies. While both methods show general alignment in delay measurements for minor movements, significant differences were observed for major movements. The MLVS, which measures delay in real-time on a per-vehicle basis, tends to report lower delays compared to the HCM method, particularly in high-volume conditions. This discrepancy may be attributed to the HCM's reliance on empirical equations that do not account for real-time fluctuations in traffic flow. These findings suggest that the MLVS offers advantages in capturing dynamic traffic conditions. The findings urge future research to broaden the scope by exploring diverse intersection types over extended periods. The study acknowledges the HCM's reliability in traffic analysis while illustrating the potential benefits of integrating machine learning techniques into future traffic management strategies for smart cities approach.