Edge Deployable Lightweight Framework for Real-Time Driver Safety System with Integrated Lane Departure and Traffic Sign Recognition

Key points

• Lane departure and traffic-sign negligence are leading contributors to road accidents, particularly in regions with non-standardized signage and limited annotated data.1
• Existing Advanced Driver Assistance Systems (ADAS) solutions are often developed as disjointed submodules, are rarely validated for real-time operation, and struggle to balance accuracy with low-latency inference for embedded deployment.1
• Our framework combines a lightweight, lane-segmentation-driven LDW, which includes a drivable-area fallback for degraded markings, with a two-stage TSD/TSR pipeline consisting of efficient sign localization followed by fine-grained classification, designed to preserve accuracy at embedded latency.1
• To support region-specific deployment under data scarcity, we curate a 35-class local traffic-sign dataset and introduce a generalized augmentation and balancing profile, expanding the corpus from 359 images to 8,750 samples.1

Sources (1)

• [1]Edge Deployable Lightweight Framework for Real-Time Driver Safety System with Integrated Lane Departure and Traffic Sign Recognition
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 18, 03:34 PM

  We address these gaps with a unified, edge-deployable ADAS framework that simultaneously performs lane-departure warning (LDW) and traffic-sign detection and recognition (TSD/TSR), and validate the proposed system on automotive-class devices.

  Lane departure and traffic-sign negligence are leading contributors to road accidents, particularly in regions with non-standardized signage and limited annotated data.

Extractive summary: sentences quoted from the sources.