[Paper Review] Automatic Estimation of the Exposure to Lateral Collision in Signalized Intersections using Video Sensors

Nicolas Saunier, Sophie Midenet|PolyPublie (École Polytechnique de Montréal)|Dec 21, 2010

Traffic and Road Safety3 references3 citations

TL;DR

This paper presents an automated method to estimate exposure to lateral collisions at signalized intersections using video sensor data, leveraging spatial occupancy and moving group detection to identify critical conflict periods. The approach achieves over 92% recall in detecting exposure to lateral collisions, demonstrating high reliability for real-time traffic safety assessment and comparison of adaptive control strategies.

ABSTRACT

Intersections constitute one of the most dangerous elements in road systems. Traffic signals remain the most common way to control traffic at high-volume intersections and offer many opportunities to apply intelligent transportation systems to make traffic more efficient and safe. This paper describes an automated method to estimate the temporal exposure of road users crossing the conflict zone to lateral collision with road users originating from a different approach. This component is part of a larger system relying on video sensors to provide queue lengths and spatial occupancy that are used for real time traffic control and monitoring. The method is evaluated on data collected during a real world experiment.

Motivation & Objective

• To develop an automated method for estimating temporal exposure to lateral collisions at signalized intersections using video sensor data.
• To enable real-time safety monitoring and evaluation of traffic control strategies by quantifying exposure to conflict events.
• To detect critical situations where a stream of vehicles crosses the conflict zone while cross-traffic is present, indicating potential lateral collision risk.
• To evaluate the performance of the method on real-world data collected during a field experiment comparing CRONOS and baseline control strategies.
• To support the analysis of safety impacts of adaptive traffic control by providing a scalable, data-driven exposure metric.

Proposed method

• The method uses video-derived spatial occupancy data to detect moving groups of road users in the conflict zone and cross-traffic approaches.
• At each second, the system identifies the origin of moving groups in conflict zones and updates cumulative durations: Z (total crossing time), X (crossing with empty cross-traffic), Y (crossing with occupied cross-traffic), and Ym (crossing with moving cross-traffic).
• The algorithm uses a rule-based detection logic: if a stream is crossing and cross-traffic is occupied, Y is incremented; if any close road user in cross-traffic is moving, Ym is incremented.
• The system processes bi-dimensional grid data representing road user presence and movement states across the intersection surface.
• The method relies on the same video sensor data used by the CRONOS real-time traffic control strategy, ensuring compatibility with existing infrastructure.
• Evaluation uses manual annotation of 10–20 minutes of peak traffic video data to validate detection accuracy for Y and Ym as binary classification tasks.

Experimental results

Research questions

• RQ1 How can exposure to lateral collision be automatically estimated using video sensor data in signalized intersections?
• RQ2 What is the accuracy of automated detection of critical conflict situations where a crossing stream encounters occupied cross-traffic?
• RQ3 How does the exposure to lateral collision vary between different traffic control strategies (e.g., CRONOS vs. baseline)?
• RQ4 To what extent does the presence of moving versus stationary vehicles in cross-traffic affect exposure metrics?
• RQ5 Can the proposed method reliably process large-scale video data for real-time safety monitoring and control evaluation?

Key findings

• The automated detection of exposure to lateral collision (Y) achieved a recall rate above 92% across all tested intersection configurations and control strategies.
• For the more complex Ym metric (involving moving cross-traffic vehicles), recall exceeded 75%, with a maximum of 100% in one configuration.
• Precision for Y detection was above 87%, indicating low false positive rates, while Ym precision ranged from 57% to 78%.
• The method’s performance was consistent across both the CRONOS adaptive strategy and the baseline time-plan strategy, suggesting robustness to control logic differences.
• The system successfully identified critical conflict periods using only occupancy data from video sensors, enabling scalable safety assessment.
• The results confirm that the method is suitable for processing large volumes of video data to support real-time traffic safety evaluation and control strategy comparison.