Advancing Cycling Safety: On-Bike Alert System Utilizing Multi-Layer Radar Point Cloud Clustering for Coarse Object...

Area of Science:

• Road safety engineering
• Sensor technology
• Machine learning for transportation

Background:

• Cyclists are vulnerable road users (VRUs) requiring enhanced protection against collisions.
• Existing safety measures often lack comprehensive detection capabilities for cyclists.
• mmWave radar offers a compact, low-power, and cost-effective sensing solution for vehicular safety.

Purpose of the Study:

• To develop and evaluate a mmWave radar-based bike safety system for real-time cyclist alerts.
• To improve obstacle detection and classification for vulnerable road users.
• To enhance overall road safety and reduce cyclist-involved incidents.

Main Methods:

• Integration of a low-power mmWave radar sensor and micro-controller on bicycles.
• Implementation of a two-level clustering algorithm with temporal projection for radar point clouds.
• Development of a coarse classification algorithm using extracted cluster features.
• Creation of the annotated RadBike dataset for system evaluation.

Main Results:

• The proposed two-level clustering achieved a v-measure score of 0.91, outperforming DBSCAN (0.88).
• Classifiers (decision trees, random forests, SVM, AdaBoost) achieved 87% accuracy in identifying four-wheeled, two-wheeled, and other objects.
• The system demonstrated effective real-time obstacle detection and classification.

Conclusions:

• The mmWave radar-based system significantly enhances cyclist safety by providing timely warnings.
• The developed clustering and classification methods are effective for obstacle detection in cycling environments.
• This technology has the potential to substantially decrease the frequency of accidents involving cyclists.