Sensor Fusion-Based Collision Detection for Bicycle Safety System

Phang, Wai Cee (2025) Sensor Fusion-Based Collision Detection for Bicycle Safety System. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Phang Wai Cee - FULL TEXT.pdf Restricted to Registered users only Download (3MB)

Abstract

Cyclists in urban environments face significant risks from rear-end collisions due to limited awareness of approaching vehicles. This thesis presents a vision-based rear collision warning system for bicycles, leveraging YOLOv8n for real-time vehicle detection with 96% accuracy for cars and trucks, and DeepSORT for robust tracking. Using monocular vision, the system estimates vehicle distances with errors below 1 meter within 10 meters and predicts trajectories via a Kalman filter to compute Time-To-Collision (TTC) with errors below 0.5 seconds. An adaptive Kalman filter used to estimate bicycle speed, achieving a mean RMSE of 0.462 m/s in simulated tests. Validated with repurposed rear car camera footage, the system correctly identifies critical scenarios (TTC < 2 seconds or distance < 2 meters), offering a cost-effective, deployable solution to enhance cyclist safety. This work advances cyclist safety by integrating state-of-the-art computer vision into a cost-effective, deployable solution, overcoming limitations of prior sensor-heavy systems.