Hybrid attention-RNN and HMM framework for reliable high-speed arterial vs. urban road classification under degraded...

Classifying vehicles as travelling on a high-speed arterial road or a low-speed urban road is important for Intelligent Transportation System (ITS), especially for applications like GNSS-based tolling. However, this task still poses a significant challenge. The traditional algorithms like Geometric, Hidden Markov Models (HMMs), and Kalman Filters fail in urban areas where GPS signals are poor, signal reflection which creates errors that confuse these algorithms. This paper proposes a novel hybrid framework that combines an attention-enhanced Recurrent Neural Network (RNN) which is trained to learn the driving or behavioral patterns like acceleration, turns on a highway vs. a service road. The RNN's output is integrated into a Hidden Markov Model (HMM), which ensures that the final path is valid. The proposed model is evaluated using the GeoLife GPS trajectory dataset. The Bi-LSTM component achieves 99.37% accuracy and 0.9909 highway F1. HMM post-processing with topology-aware spatial emissions reduces physically impossible singleton road-type transitions by 9.4%, improving sequence coherence for real-world ITS deployment. Stratified OSM validation on 250 GPS points confirms 98.4% label agreement with ground-truth road-type annotations, and degraded GPS conditions are simulated via Gaussian and multipath noise to reflect urban-canyon environments.