2025 PSR UTC Annual Congress

Enhancing Surrogate Safety Measures for Active Transportation: A Multi-Metric Approach to Near-Miss Detection and Risk Assessment - As transportation systems professionals strive to provide more sustainable options of travel, there is a growing need to understand the safety risk of active travelers such as cyclists, pedestrians, and users of shared micro-mobility services. A lack of active transportation crash data challenges the ability to identify problematic areas requiring improvements. Surrogate Safety Measures (SSM) are traditionally used to substitute gaps in crash data for safety interventions, however, the existing research is centered around motorized vehicles. Metrics such as Post Encroachment Time (PET), Time to Collision (TTC) don’t capture the less predictable movements and behavioral dynamics of active transportation users. Existing metrics may overlook key factors that contribute to near-miss events among vulnerable road users.

This study conceptualizes the effectiveness of different combinations of surrogate safety measures [Time Difference to Point of Intersection (TDPI), Speed-Distance Profile, Distance between Stop Point and Pedestrian (DSPP), Conflict Index (CI), and Deceleration Rate to Avoid Crash (DRAC)] in detecting near-miss events among active transportation user interactions [bicycle-to-bicycle, bicycle-to-pedestrian, bicycle-to-e-bike, etc.]. To do this, we categorize interactions by mode type, turning movements, merging, and platooning, then assess the trajectories, speeds, position data, and kinetic potential of the interactions to evaluate the applicability of existing metrics and investigate new measures. Different metrics may be effective at identifying certain types of near-misses while being less effective for others. Moreover, validating the accuracy of the metrics is difficult without concurrent active transportation crash data. However, a combination of controlled experimentation along with natural data collection might help improve the applicability of these metrics.

The study starts with series of micro-experiments where, simulated near-miss incidents across various modalities are videotaped, with participants providing a subjective rating of risk (low, medium, or high) immediately after each incident through a survey. Different metrics are then used to analyze the concordance between human perception and metrics assigned near-misses. This process is to refine the metrics and identifies conceptual gaps before transitioning to real-time data collection and model development. The data is collected using commercial computer vision-integrated cameras on the University of California, Davis campus. With the highest mode share of active transportation among university campuses in the US, UC Davis provides a unique data-rich environment to study the events.

After the data collection, the model cycles through various combinations of SSMs to achieve the best fit for near-miss detection for all selected categories. The study proposes a risk scale for interactions instead of depending on fixed threshold values for presenting the results. The flagged interactions are further examined by trained observers to validate the detection process and evaluate the correlation between identified near-misses and perceived risk. If successful, these findings will contribute to surrogate safety measures tailored specifically for assessing active transportation safety. The future research can aim to confirm these surrogate measures against crash data in different spatial and temporal contexts to further enhance their validity and usefulness.