EVALUATION OF INTERSECTION SAFETY USING SPEED-BASED INDICATORS IN MICROSIMULATION

Abstract

Traffic safety is an important area of transportation engineering. In the developing world, with increases in population, the number of vehicles is increasing tremendously. Traffic safety on roads has become a major concern even with advancements in technology and infrastructure. Presently, most traffic safety assessment and prediction related work are based on the use of historical accident data that has known drawbacks related to the quality and coverage of data especially in developing countries like India. For the assessment of roadway solutions in the future, it is impractical and unethical to wait for accidents to occur before being able to draw statistically sound conclusions regarding safety impact. Hence, it is imperative to adopt safety evaluation technique which is proactive in nature. The main advantage associated with these safety indicators is that they occur considerably more frequently than accidents, thereby implying an efficient and more statistically reliable proximal measure of traffic safety. The proximal indicators that are proposed as the best are time to collision, post-encroachment time, deceleration rate, and speed differential. But, despite decades of conceptual development and widespread application, there are still some disputes on whether a traffic conflict should be in fact defined and separated as critical or severe conflict from other non-severe conflict events. Several researchers have defined different threshold values for proximal safety indicators to identify conflicts in a certain situation. However, use of these threshold values is in doubt in the highly heterogeneous Indian traffic condition? Another problem that limits the application of the time-based indicators (PET, TTC, etc.) is that several combinations of speed and distance measures can produce same situations and unable to describe the severity of conflict situation. Therefore, to overcome the problems related with time-based indicators, individual use of surrogate safety measures which is not able to provide enough information to describe a critical conflict situation, a new modified surrogate safety function called “Criticality Index Function” is defined which incorporates several proximal safety indicators in a functional form. Further, Gaussian Mixture Model which is an unsupervised Machine Learning technique used to label the unlabeled data in the large dataset has been used to identify critical conflicts. The objective of this study is to identify critical conflicts out of iv total vehicular interactions, with the help of unsupervised machine learning without using the concept of the threshold value of surrogate measures. Conflict points generated with Criticality Index Function are categorized in the three categories “severe”, “mild-severe” and “non-severe” with GMM (Gaussian Mixture Model). With this proposed methodology of safety assessment, it is possible to overcome the demerits of using a certain threshold of surrogate measure which is sensitive to a geometric feature of roadway and traffic condition. The results of this research are founded out to be encouraging as it is feasible to use the proposed automated method for the identification of critical conflicts for any geometric and traffic condition, which has the potential to convert to accidents. Finally, various possible alternatives like a roundabout, grade-separation (provision of flyover), application of speed breaker at the minor road are applied to optimize the performance of intersections by comparing Criticality Index Function per vehicle (CIF/veh.) from a safety perspective.