42 / 2025-03-28 17:50:24

Lightweight Multi-Scale Feature Extraction Network for Efficient Driver Gaze Tracking

Driver gaze estimation,lightweight network,attention mechanism,multiscale feature extraction

We present a lightweight multiscale feature extraction network designed for efficient and real-time driver gaze estimation in resource-constrained environments, such as in-vehicle systems. The proposed architecture integrates three key components: a residual backbone built with depthwise separable convolutions for multiscale feature extraction, a hybrid local-global attention mechanism to capture both spatial detail and contextual dependencies, and a dynamic feature fusion module that adaptively reweights features based on contextual relevance. Extensive experiments on the Gaze360 and IVGaze datasets demonstrate that our model achieves competitive accuracy with significantly fewer parameters and lower computational cost compared to existing high-performance methods. These results highlight the model’s suitability for real-world deployment in intelligent transportation and driver monitoring systems.