Driver distraction detection and recognition using RGB-D sensor

Craye, Celine; Karray, Fakhri · 2015 · arXiv

archive: archived pipeline: cataloged verified

Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the critical safety issue of distracted driving, which contributes significantly to traffic accidents and fatalities. The authors propose a computer vision-based system to not only detect if a driver is distracted but also recognize the specific type of distraction (e.g., phone use, eating, texting). While existing methods often focus on fatigue or general inattention, this work specifically targets distraction recognition using an active RGB-D sensor (Microsoft Kinect) to capture both color and depth data. The system is designed to be non-intrusive and suitable for integration into intelligent transportation systems. The methodology employs four independent feature extraction modules: arm position estimation, eye behavior (gaze and blinking), head orientation, and facial expressions. Arm position is determined using depth map segmentation and principal component analysis on contour features, classified via AdaBoost. Eye behavior utilizes color data with filters for iris localization and an SVM classifier for eye closure detection. Head pose and facial animation units are derived from the Kinect SDK’s face tracking algorithm. These features are fused using two classification strategies: an AdaBoost classifier with temporal smoothing and a Hidden Markov Model (HMM). The system was evaluated using a driving simulator with eight drivers of varying demographics, recording over eight hours of data across five distinct tasks: normal driving, making a phone call, drinking, sending an SMS, and looking at an object inside the vehicle. The results demonstrate high efficacy in both detection and recognition. The system achieved an average accuracy of 85.05% for AdaBoost and 84.78% for HMM in recognizing the specific type of distraction. For binary distraction detection (distracted vs. normal), accuracy reached approximately 90% for both classifiers. Analysis of individual modules revealed that arm position features were the most discriminative overall, while head orientation was particularly effective for detecting drinking and texting, and eye behavior aided in identifying texting and normal driving. The HMM performed comparably to AdaBoost but showed instability for some drivers, likely due to the high dimensionality of features and the variability of non-cooperative driver actions. Phone calls and normal driving were detected with the highest precision, while object distraction was the most challenging due to its similarity to normal driving behaviors. The significance of this work lies in its ability to provide granular information about driver behavior, which can enhance intelligent vehicle assistance systems beyond simple alert mechanisms. By identifying the specific nature of the distraction, the system can offer more contextual safety interventions. The modular design allows components to be repurposed for other inferences, such as fatigue detection. The study confirms that RGB-D sensors combined with robust computer vision techniques can effectively monitor driver attention in real-world conditions, offering a viable, non-intrusive solution for improving road safety.

Key finding

A Kinect-based RGB-D system with AdaBoost and HMM classifiers can detect driver distraction states from visual features including eye gaze, arm position, head pose, and hand movement.

Methodology

simulator

Sample size: 8

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via discover_arxiv on 2026-05-04 (4 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified_with_issues.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• distraction detection algorithms
• drowsiness detection algorithms
• visual
• gaze based attention detection

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data
• Methodological Resource: tool software
• Theoretical Contribution: conceptual framework