Detecting & Predicting Micro-Mobility Riders' Behaviors

Overview

Micro-mobility (sharing bikes, e-scooters, etc.) has become a vital mode of urban transportation, but it has also introduced a rise in traffic accidents. This project focuses on detecting and predicting riding behavior using consumer devices such as smartphones and earbuds to enable active protection—issuing timely warnings and interventions before accidents occur.

Approach

• HeadMon / HeadMon+: Predicting riding maneuvers from head dynamics captured by helmet/earbud-mounted IMUs. HeadMon+ incorporates a domain adversarial network for robust adaptation to new riders and conditions
• RideGuard: Leveraging a smartphone's built-in IMU with a dual-stream CNN to predict steering maneuvers from micro-mobility body dynamics in both time and frequency domains
• DoubleCheck: Real-time single-handed cycling detection and distraction activity recognition using a handlebar-mounted smartphone's IMU
• HeadSense: Visual search monitoring and distracted behavior detection for bicycle riders using a helmet-mounted 9-axis IMU
• RideStyle: Riding style representation from head-body dynamics via adversarial learning

Results

• HeadMon+ (IEEE TMC 2025): 94% overall precision with a 4-second prediction time gap; domain adaptation enables single-time pre-training for new riders and conditions
• RideGuard (IEEE ICDCS 2024): 94%+ precision with a 5-second prediction time gap using a 3-second detection window; validated with 20 participants in real traffic
• HeadMon (IEEE PerCom 2023): 85%+ precision with a 4-second prediction time gap; maintains high accuracy even at low sampling rates
• DoubleCheck: F1-score of 0.96 for single-hand detection with 22 participants on asphalt and pavement

Significance

By leveraging low-cost consumer devices, this project enables feasible and scalable deployment of active safety systems for micro-mobility riders. The rapid response capabilities of these systems contribute to enhancing safe riding practices without requiring specialized infrastructure.