Real-Time Motorcycle Hazard Detection and Warning System: A Multi-Stage Image Enhancement and YOLOv8 Pipeline

Motorcycle accidents in India represent a critical public safety concern, driven by a confluence of factors such as substandard road infrastructure, unpredictable weather conditions, and human errors [1]. These incidents result in significant loss of life and economic burdens, necessitating innovative technological interventions. This paper introduces an advanced motorbike safety system designed to mitigate these risks through real-time image processing and object detection. The proposed system leverages a multi-stage pipeline that integrates dynamic condition classification, adaptive image enhancement, and precise hazard detection. Initially, a ResNet-18 model classifies environmental conditions such as rain, fog, or night time, enabling context-sensitive processing. Subsequently, a combination of U-Net architecture and Contrast Limited Adaptive Histogram Equalization (CLAHE) enhances image visibility under adverse conditions, achieving a Structural Similarity Index (SSIM) of 0.83. For hazard detection, a custom-trained YOLOv8 model identifies critical obstacles within a 5-metre danger zone, including potholes (average confidence: 0.64), speed bumps (0.61), and nearby vehicles (0.68) [2]. The system was trained on a comprehensive dataset of over 5,000 images capturing diverse Indian road scenarios, ensuring robust performance across varied environments. Implemented on embedded hardware, the system delivers real-time processing at 30 frames per second (FPS), making it both efficient and practical. By providing immediate alerts for detected hazards, this solution offers a cost-effective and scalable approach to improving motorbike safety. Beyond its immediate application, the system holds promise for integration into broader intelligent transportation frameworks, contributing to safer and smarter mobility solutions in challenging driving contexts.