Bayesian Destination Inference for Predictive Congestion Prevention in Indian Urban Corridors: The PTIS v2.0 Architecture

Urban traffic management systems in Indian cities face two compounding challenges absent from Western deployments: two-wheelers constitute 62–65% of the vehicle fleet and behave as compressible fluid rather than discrete agents, and navigation applications (Google Maps, Waze) with 85%+ smartphone penetration route independently of infrastructure signals, creating secondary congestion on designated diversion routes. We present PTIS v2.0 (Predictive Traffic Intelligence System), a four-layer predictive architecture for the Bengaluru Outer Ring Road corridor (Silk Board → Whitefield, 18 km, 6 junctions) that addresses both challenges through a novel integration of Bayesian destination inference and dynamic connector-road activation. PTIS's primary contribution is corridor-level Bayesian destination inference from junction checkpoint observations: each junction a vehicle passes without turning eliminates destination candidates, building a posterior that achieves 65–88% confidence 2–3 junctions before the destination. This confidence triggers smart link activation — dynamic opening of existing connector roads — before congestion forms rather than after. Secondary contributions include: (1) a real-time Lighthill-Whitham-Richards (LWR) fluid model solved with the Godunov scheme for two-wheeler density estimation; (2) a PID closed-loop compliance controller eliminating the fixed-compliance assumption used by all deployed systems; (3) a joint navigation app co-optimisation framework via Google Road Manager API preventing secondary bottleneck formation. Simulation on the Silk Board to Whitefield corridor shows 43% congestion reduction, 8.2 min/vehicle travel time savings, and Rs. 374 Crore/yr (approx. USD 45M/yr) annual economic benefit at Rs. 23.74 Crore capex. No currently deployed traffic management system in India combines these four mechanisms.