Existing urban navigation systems are based on shortest-path algorithms and ignore safety-related hazards such as road defects, traffic conditions, and current environmental influences. In this paper, we propose SafeRoute AI, an AI-based, risk-aware, crowdsourced smart navigation system that integrates YOLOv8-based real-time road hazard detection, crowdsourced user reports, a trust-aware probabilistic validation method, and a multi factor risk-aware routing approach. We modify the Dijkstra cost function to include hazard intensity, report frequency, and traffic conditions as cost factors in addition to distance. Ablation studies show how each component contributes to overall system performance. The results indicate that the proposed system achieves 85.9% mAP for hazard detection, 94.2% trust-validation accuracy, and produces routes that are 47% safer than shortest-path routing while adding only 6.3% distance overhead.

—Road Hazard Detection; Risk-Aware Routing; Crowdsourcing; YOLOv8; Deep Learning; Intelligent Transportation Systems; Trust Validation