Assessing Vulnerable Road User Safety at Roundabouts using Video Analytics

Abstract

Roundabouts are typically considered a safer type of intersection. By physically deflecting the approaching traffic, the roundabouts force motorised vehicles to slow down, according them the opportunity to react to any potential conflicts more judiciously. They also stagger the conflict points and zones around a larger physical area helping to prevent multiple conflicts occurring simultaneously. For these reasons they are often recommended as a safety countermeasure in certain situations. However, roundabouts are not the safest type of intersections for vulnerable road users (VRUs). Due to their larger footprint, they are often not convenient to cross for pedestrians and bicyclists. As vehicles are not required to come to a complete stop, the likelihood of low intensity conflicts with VRUs are usually high, especially in the absence of adequate refuge areas. In the City of Gold Coast, 136 crashes occurred at roundabouts in 2021, of which 21 (15%) involved either pedestrians or bicyclists. Unfortunately, due to their small sample sizes and other statistical problems, traffic crashes involving VRUs are not the most reliable indicators of safety issues. Further, the crash data from the last 3 years with external impacts of COVID means the data is not representative of the underlying characteristic crash generating process. Rather than exclusively using crash data, using traffic conflict techniques, which rely on traffic conflicts as the proactive indicator of safety issues at a site, are a more suitable alternative for targeted safety assessments. An Advanced Mobility Analytics Group (AMAG) study used video-based traffic conflict analytics to analyse the safety of three roundabouts in the City of Gold Coast. The traffic movements were observed using high-mast video cameras installed at the study intersections for one week, from 6 am to 6 pm each day, at the study sites. The recorded movements were analysed using the AMAG's cloud-based SMART SafetyTM platform that uses deep learning methods to extract high-quality road user trajectories, flows, speeds, and critical safety metrics. The safety metrics included the number and proportions of violating road users as well as conflicts measured using the Time-to-Collision (TTC) and Post-Encroachment Time (PET) indicators. If either of the two indicators fell below 1.5 s at any time during a pedestrian-vehicle interaction, then that interaction was labelled a conflict. The violations captured by the platform included speeding (going 5 km/h above the posted speed limit) by motorised vehicles and spatial violations by pedestrians, where pedestrians crossed the street at locations other than the designated crosswalk. The safety improvements were measured by comparing the conflict intensity indices and conflict rates (both absolute and corrected for exposure) of pedestrians, bicyclists, and e-scooter riders at the three roundabouts. The results of the study can help Councils and transport authorities to prioritise and design the safety treatments at roundabouts to enhance VRU safety and achieve the overarching Vision Zero goals. Moreover, the findings of this study emphasise the effectiveness of video analytics in proactive safety evaluations overcoming the arguably unethical requirement of waiting for crashes to accrue.