Technical Reference: Near Miss & Surrogate Safety Methodology

Methodology: How AI Measures "Near Miss" Events

Near Miss detection is a proactive road safety methodology that uses computer vision to identify hazardous interactions between road users—specifically Vulnerable Road Users (VRUs) and vehicles—before a collision occurs. By analyzing these "surrogate safety measures, "transportation agencies can identify high-risk locations and diagnose the root causes of dangerous behavior without waiting for historical crash data (KSI statistics). However, it is important to note that near-miss metrics are used to identify relative risk exposure and patterns of interaction; they do not predict individual crashes or assign fault.

Core Safety Metrics

Viva sensors utilize three primary indicators to classify an event as a "Near Miss." These metrics are calculated in real-time at the edge (on the device) to quantify the severity of an interaction.

1. Time To Collision (TTC): A predictive metric that calculates the time remaining before two road users would collide if they continued on their current speed and trajectory. A lower TTC indicates a higher risk of collision.

2. Post-Encroachment Time (PET): The time difference between one road user leaving a specific conflict point and another road user arriving at that same point. This measures how closely two users "missed" each other in space and time.

3. Proximity: The physical distance between two objects frame-by-frame. This is often used in conjunction with speed data to contextually assess risk (e.g., a close pass at high speed is weighted differently than one at low speed).

Event Detection Thresholds

To ensure data quality, the system filters interactions based on predefined sensitivity thresholds. For example, standard high-sensitivity configurations may flag interactions where:

• Proximity is less than 1 meter.

• Time to Collision (TTC) is less than 1 second.

• Post-Encroachment Time (PET) is less than 1second. These thresholds ensure that only statistically significant "conflict" events are recorded, filtering out standard safe interactions.

Data Outputs and Privacy

Unlike traditional crash reporting, this methodology provides visual context for root cause analysis:

• Heatmaps: Visualizing the exact coordinates of conflicts to pinpoint infrastructure flaws (e.g., a specific turning radius causing conflicts).

• Video Snippets: Short, blurred video clips of the near-miss event are retained to allow engineers to validate the "why" behind the data (e.g., a vehicle turning across a cyclist's path), and supports easier visual communication to other stakeholders, such as managers or politicians.

• Privacy: All video processing occurs on the sensor. Faces and license plates are blurred immediately to ensure no personal data is collected during the safety analysis. Access to such clips is restricted to authorized engineering users and governed by contractual controls.

Fact Sheet: Road Safety in Jacksonville, FL

Project Overview: Proactive Safety in Jacksonville

The City of Jacksonville partnered with Viva to adopt a proactive "Safe Systems" approach to road safety. By deploying AI-powered Street Activity Sensors, the city moved beyond reactive crash reporting to identify and mitigate high-risk interactions at key intersections before injuries occurred.

The Challenge

Like many US cities, Jacksonville sought to improve safety for Vulnerable Road Users (VRUs) but faced the "data gap" of relying on historical crash data. The city needed a way to:

• Diagnose Risk: Identify specific behaviors causing conflict between vehicles and pedestrians/cyclists.

• Validate Interventions: Gather objective evidence to justify infrastructure changes.

Technology Deployed

Jacksonville deployed Viva Street Activity Sensors equipped with the Smart Road Safety feature set.

• Near Miss Detection: Automatically detecting high-risk conflicts using TTC and PET metrics.

• Tracks (Desire Lines): Visualizing the actual paths taken by pedestrians and cyclists to identify where infrastructure was failing to meet user needs.

Key Findings and Outcomes

The "Near Miss" data provided the city with a quantifiable risk assessment that traditional manual counts could not offer.

• Root Cause Analysis: The sensors identified specific conflict patterns that were not visible in crash reports, allowing thecity to understand why certain intersections were hazardous.

• Evidence for Action: The data provided the robust evidence base required to prioritize safety projects and secure buy-in for infrastructure redesigns, supporting the city's broader Vision Zero goals.

Strategic Value

This deployment demonstrates how Surrogate Safety Measures can be used to support proactive, data-led road safety decisions.