Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Contextual Analysis of Driver-Facing and Dual-View Video Inputs for Distraction Detection in Naturalistic Driving Environments

Anthony Dontoh, Stephanie Ivey, Armstrong Aboah

TL;DR

This paper addresses whether incorporating road-facing context improves distraction detection in naturalistic driving. It systematically compares driver-only versus stacked dual-view inputs across SlowFast-R50, X3D-M, and SlowOnly-R50 architectures on real-world video data. The results show that dual-view fusion yields architecture-dependent outcomes: SlowOnly-R50 benefits with dual-view (≈9.8 percentage points), SlowFast-R50 can deteriorate (≈7.2 points), and X3D-M performs best with driver-only inputs (~55.3% accuracy). The findings emphasize that naive input stacking can hinder performance, underscoring the need for fusion-aware architectures that effectively leverage multiple views in context-rich driver monitoring systems.

Abstract

Despite increasing interest in computer vision-based distracted driving detection, most existing models rely exclusively on driver-facing views and overlook crucial environmental context that influences driving behavior. This study investigates whether incorporating road-facing views alongside driver-facing footage improves distraction detection accuracy in naturalistic driving conditions. Using synchronized dual-camera recordings from real-world driving, we benchmark three leading spatiotemporal action recognition architectures: SlowFast-R50, X3D-M, and SlowOnly-R50. Each model is evaluated under two input configurations: driver-only and stacked dual-view. Results show that while contextual inputs can improve detection in certain models, performance gains depend strongly on the underlying architecture. The single-pathway SlowOnly model achieved a 9.8 percent improvement with dual-view inputs, while the dual-pathway SlowFast model experienced a 7.2 percent drop in accuracy due to representational conflicts. These findings suggest that simply adding visual context is not sufficient and may lead to interference unless the architecture is specifically designed to support multi-view integration. This study presents one of the first systematic comparisons of single- and dual-view distraction detection models using naturalistic driving data and underscores the importance of fusion-aware design for future multimodal driver monitoring systems.

A Contextual Analysis of Driver-Facing and Dual-View Video Inputs for Distraction Detection in Naturalistic Driving Environments

TL;DR

This paper addresses whether incorporating road-facing context improves distraction detection in naturalistic driving. It systematically compares driver-only versus stacked dual-view inputs across SlowFast-R50, X3D-M, and SlowOnly-R50 architectures on real-world video data. The results show that dual-view fusion yields architecture-dependent outcomes: SlowOnly-R50 benefits with dual-view (≈9.8 percentage points), SlowFast-R50 can deteriorate (≈7.2 points), and X3D-M performs best with driver-only inputs (~55.3% accuracy). The findings emphasize that naive input stacking can hinder performance, underscoring the need for fusion-aware architectures that effectively leverage multiple views in context-rich driver monitoring systems.

Abstract

Despite increasing interest in computer vision-based distracted driving detection, most existing models rely exclusively on driver-facing views and overlook crucial environmental context that influences driving behavior. This study investigates whether incorporating road-facing views alongside driver-facing footage improves distraction detection accuracy in naturalistic driving conditions. Using synchronized dual-camera recordings from real-world driving, we benchmark three leading spatiotemporal action recognition architectures: SlowFast-R50, X3D-M, and SlowOnly-R50. Each model is evaluated under two input configurations: driver-only and stacked dual-view. Results show that while contextual inputs can improve detection in certain models, performance gains depend strongly on the underlying architecture. The single-pathway SlowOnly model achieved a 9.8 percent improvement with dual-view inputs, while the dual-pathway SlowFast model experienced a 7.2 percent drop in accuracy due to representational conflicts. These findings suggest that simply adding visual context is not sufficient and may lead to interference unless the architecture is specifically designed to support multi-view integration. This study presents one of the first systematic comparisons of single- and dual-view distraction detection models using naturalistic driving data and underscores the importance of fusion-aware design for future multimodal driver monitoring systems.
Paper Structure (18 sections, 4 figures, 1 table)

This paper contains 18 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Naturalistic Driving Datasets
  4. Single-View Detection Approaches
  5. Methodology
  6. Dataset Description and Preparation
  7. Model Architectures
  8. SlowFast Networks
  9. X3D Networks
  10. SlowOnly Networks
  11. Training Strategy and Hyperparameters
  12. Evaluation Metrics
  13. Results and Discussion
  14. Model Performance Overview
  15. Per-Class Analysis
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Architectural block diagram of the experimental setup. Both dual-view (driver + road) and single-view (driver only) inputs are evaluated using three backbone action recognition models (SlowFast, X3D-M, and SlowOnly), with outputs classified into distraction categories.
  • Figure 2: Distribution of distraction classes across training, validation, and test sets.
  • Figure 3: Comparison of classification accuracy across model architectures using driver-only versus stacked dual-view inputs (top), and corresponding accuracy differences (bottom).
  • Figure 4: Confusion matrices comparing classification results across model architectures using driver-only (left) and stacked dual-view inputs (right).