DNC-IMM: Early Lane-Change Intention Recognition via Neural Calibration Based on Driving Context Information
A paper proposes a Dual Neural-Calibrated Interacting Multiple Model (DNC-IMM) for early lane-change intention recognition. It encodes driving-context information with a neural network to calibrate transition probabilities and measurement likelihoods. Experiments on the highD dataset show reliable recognition before lane crossing, with strong performance at 2-3 second prediction horizons.
The paper introduces DNC-IMM, which improves adaptability to driving context while preserving the probabilistic structure and interpretability of a conventional IMM. The method uses a neural network to calibrate both the transition-probability matrix and measurement likelihoods based on target-vehicle motion, gaps to surrounding vehicles, and relative velocities. The final intention is determined from the calibrated IMM mode posterior rather than a separate classifier. Experiments on the highD dataset demonstrate reliable lane-change intention recognition before lane crossing, particularly at earlier 2-3 second horizons.
The approach integrates neural calibration into the IMM framework, adjusting transition probabilities and measurement likelihoods using contextual features. This maintains probabilistic interpretability while enhancing adaptability. Evaluation on highD shows improved early recognition, especially at 2-3 seconds before lane crossing, suggesting the calibrated posterior better captures evolving driver intent.
Early lane-change intention recognition can improve proactive decision-making in autonomous driving and advanced driver assistance systems. The method's strong performance at 2-3 second horizons is relevant for safety-critical applications where earlier anticipation allows smoother and safer maneuvers.
Improved early intention recognition could enhance the safety and user experience of autonomous driving systems, potentially reducing accidents and enabling more natural vehicle behavior. This may increase the value proposition for OEMs and ADAS suppliers.
Potential next steps include validation on additional datasets, real-time implementation, and integration with downstream planning modules. Observing whether the approach generalizes to other driving scenarios or is adopted in commercial ADAS stacks would be a key signal.