Mobile sensing discovery of when where and why vulnerable road users break traffic rules

Abstract This study performed a large-scale analysis of vulnerable road user (VRU) violation behaviors in Beijing using a novel Rotating Mobile Monitoring method with machine learning. Across four seasons, we processed 367,076 street-view images and identified 20,616 violations. Private e-bike users were the primary violators (52.9%), with not wearing a helmet being the most common infraction (11,714 instances). These behaviors exhibited clear temporal patterns, peaking in spring and during the afternoon. The built environment was a key predictor, with building and commercial activity within a 150-meter buffer correlating with multiple violation types. This research quantifies predictable risk patterns, directly linking violation hotspots to features like commercial density within a 15 minute life circle (150 m radius). This evidence enables targeted interventions for specific user groups, times, and locations, providing a data-driven path towards safer urban transport. The scalable methodology also presents a practical tool for ongoing urban safety diagnostics.