International Journal of Science, Engineering and TechnologyAuthors: Dr. Pankaj Malik, Yash Upadhyay, Swaraj Singh Rajput, Anuj Patidar, Taher Fakhri
Abstract: Underbody collision is a vehicle-specific safety problem encountered while traversing potholes, speed breakers, road humps, deep depressions, rocks, and uneven or off-road surfaces. Conventional vision-based advanced driver assistance systems (ADAS) can detect road anomalies, but anomaly detection alone does not determine whether a particular vehicle will experience underbody contact. This paper presents an Intelligent Vehicle-Specific Underbody Collision Prediction (IV-UCP) framework that integrates forward-camera road-obstacle detection, road-geometry estimation, vehicle geometry, vehicle dynamics, and a physics-guided minimum-clearance model. The framework represents road geometry, vehicle characteristics, and dynamic state in a multimodal feature vector and uses machine learning to estimate collision probability and risk level. A conservative safe-speed module, cross-vehicle evaluation protocol, ablation analysis, explainable AI, and real-time deployment evaluation are also incorporated. The physical criterion C_min = min_x[Z_u(x)-Z_r(x)] is used as an interpretable clearance feature, while uncertainty and ground-truth quality are treated as important experimental considerations. The numerical values currently retained in the manuscript are explicitly identified as tentative or target values where verified experimental measurements are not available; they must be replaced by measured results before making empirical performance claims. The framework therefore provides a structured path from road-anomaly recognition to vehicle-specific underbody collision-risk prediction and driver assistance.
