FNI continues development of video conflict analysis capabilities
Road safety analyses and the development of safety performance functions (SPFs) have traditionally relied upon historical collision data, however the use of historical collision data causes a delay until collisions occur before safety conclusions can be made. Another challenge with using historical collision data is underreporting.
Serious conflicts or near misses occur more frequently than collisions. Unlike collisions, conflicts are not reported, observed, or analyzed to the same degree as collisions. There is inherent subjectivity in safety analysis of safety surrogates (e.g. near misses, deceleration rates, and speed differentials) when data collection is restricted to on-site observation or manual review of video, particularly when assessing the severity of a conflict. Furthermore, this analysis is both time consuming and tedious.
The use of artificial intelligence (AI) and machine learning enables automated conflict analysis, saving time and increasing accuracy. Similarly, computer-aided video conflict analysis broadens the scope of possible conflict types to be considered.
Analyzing video data of road facilities for detection of conflicts presents an unprecedented opportunity to harness as much data as possible that will provide insight into possible conflict situations and make it easier to suggest effective treatment strategies. One such example is conflict analysis at a new intersection where no collisions have been recorded despite strong indications from the public that safety issues are present. Video conflict analysis could help identify safety issues at the site, leading to effective treatment selection and reduced probabilities of fatal and injury collisions.
Some conflict scenarios to consider include rear end, right angle, side swipe, and pedestrian-related conflicts. The severity of these conflicts may be measured using different safety indicators, such as collision course angle, velocity angle, distance/proximity, speed differential, collision probability, time to collision (TTC), and gap time (also referred to as post encroachment time or PET).
A number of time-based indicators such as TTC are obtained by predicting the future position of road users from their initial positions, given their speed and orientation. These projections are based on motion prediction algorithms such as constant velocity, normal adaptation (modifies constant velocity to adapt for driver behavior), and motion pattern learning. The precision of the calculated indicators is largely affected by the selection of the motion prediction algorithm.
Fireseeds North Infrastructure continues to develop video conflict analysis software with support from the National Research Council Industrial Research Assistance Program. This software enables road agencies to assess and quantify near misses at facilities on their road network. In turn, this information supports rapid before and after studies, treatment evaluations, conflict-based network screening, and improved diagnostics for in-service road safety reviews.