Road Width Estimation using Multi-Camera Deep Learning Framework in Unmarked Indian Roads: Quantitative Modeling for Economic Prediction

The accurate estimation of road widths on unmarked and unstructured roads is critical for Advanced Driver Assistance Systems (ADAS), autonomous navigation, and intelligent transportation planning. In Indian driving conditions, irregular boundaries and missing lane markings make conventional lane detection ineffective. We propose a multi-camera framework integrating a dashboard monocular camera with dual Outside Rear View Mirror (ORVM) cameras for dynamic road-width estimation. The scheme employs deep learning segmentation to extract road boundaries and fuses monocular depth and stereo disparity using a Kalman-based dynamic model. From a quantitative perspective, the recursive Kalman fusion functions analogously to dynamic optimization and equilibrium modeling in 'Quantitative Economics, allowing predictive analysis of traffic density, flow efficiency, and infrastructure utilization. Trained on a diverse urban–rural Indian road dataset, the model achieves high estimation accuracy and predictive stability, demonstrating improved performance over single-camera baselines and enabling data-driven optimization for safe, efficient transportation systems.