Computer vision and statistical insights into cycling near miss dynamics

Area of Science:

• Urban mobility and transportation safety
• Computer vision and artificial intelligence
• Human-computer interaction

Background:

• Urban cycling offers health and environmental benefits but faces safety concerns.
• Near-miss incidents, though not always resulting in injury, are common and indicate cyclist risk.
• Manual analysis of sensor data from naturalistic cycling studies is time-consuming and limits research scope.

Purpose of the Study:

• To develop an automated computer vision framework for analyzing cyclist near-miss events.
• To identify and assess the statistical significance of various risk factors contributing to near misses.
• To enhance the efficiency and expand the scope of naturalistic cycling safety studies.

Main Methods:

• Utilized a novel computer vision framework for automated video analysis of cyclist near-misses.
• Implemented Granger causality analysis to determine the temporal relationship between risk factors and near misses.
• Incorporated varied time lags to examine the dynamic influence of factors such as glare, vehicle, and pedestrian presence.

Main Results:

• The framework successfully automates the detection of multiple risk factors in near-miss scenarios.
• Identified significant correlations between factors like glare, vehicle, and pedestrian proximity and near-miss occurrences.
• Quantified the statistical significance of these risk factors, providing insights into their impact on cyclist safety.

Conclusions:

• The developed framework significantly enhances the efficiency and scope of naturalistic cycling studies.
• Automated analysis of near-miss data provides a deeper understanding of urban cyclist risks.
• Future integration with edge AI could enable real-time risk detection and prevention for cyclists.