DualScope: Capturing Critical Spatial and Temporal Cues for Distracted Driving Activity Recognition

Accurately recognizing distracted driving activities in realworld scenarios is essential for improving road and pedestrian safety. However, existing approaches are prone to attend to irrelevant scene contexts and are susceptible to interference from redundant frames, compromising their robustness in complex driving environments. To overcome these limitations, we propose DualScope, a novel framework that captures behaviorally critical information from spatial and temporal perspectives. In the Spatial Scope, we introduce a Synergistic Behavior-Centric Distillation mechanism that leverages two key information sources: (1) position-aware knowledge derived from the SAM model that enhances the perception of critical regions and their semantic interaction structures and (2) fine-grained visual details obtained from cropped key regions that improve the model's ability to capture detailed patterns in behavior-relevant areas. In the Temporal Scope, we present the Saliency-Aware Fine-to-Coarse Temporal Modeling module comprising three components: a Fine-Grained Motion Encoder for capturing local inter-frame dependencies, a Dynamic Difference Extractor for extracting salient motion dynamics, and a Saliency-Aware Temporal Pyramid Mamba for integrating these features to enable multiscale temporal modeling. This design effectively captures short-term motions and long-term behavioral patterns. Furthermore, incorporating salient dynamics enhances the model's focus on substantial behavioral variations. Extensive experiments on seven publicly available distracted driving activity recognition datasets demonstrate that DualScope consistently outperforms state-of-the-art methods, validating its effectiveness in capturing behavioral cues across spatial and temporal dimensions.