Owl and Lizard: Patterns of Head Pose and Eye Pose in Driver Gaze Classification

Fridman, Lex; Lee, Joonbum; Reimer, Bryan; Victor, Trent · 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 challenge of accurately classifying driver gaze from uncalibrated monocular video, a critical capability for improving vehicle interfaces and Advanced Driver Assistance Systems. The authors investigate two primary questions: first, the extent to which adding eye pose estimation improves gaze classification accuracy compared to using head pose alone; and second, whether individual-specific gaze strategies correlate with this improvement. The study is motivated by the difficulty of gaze tracking in driving environments due to varying lighting, occlusions, and vehicle vibration, and the need to understand when eye movement data provides significant value over head movement data. The researchers evaluated data from an on-road study involving 40 drivers performing secondary tasks in instrumented vehicles. The dataset consisted of over 1.3 million annotated video frames, manually labeled into six gaze regions (road, center stack, instrument cluster, rearview mirror, left, and right). The proposed pipeline utilizes face detection via Histogram of Oriented Gradients, face alignment using a cascade of regressors for 68 facial landmarks, and pupil detection using an adaptive Cumulative Distribution Function (CDF) method. Features were normalized based on the bounding box of the eyes and nose, and a random forest classifier was employed to predict gaze regions. The system filtered for high-confidence decisions, resulting in an effective decision rate of 1.3 Hz. The results demonstrate that incorporating eye pose alongside head pose increases overall gaze classification accuracy by 5.4%, raising it from 89.2% to 94.6%. The most significant accuracy gains occurred for the "center stack" region, suggesting drivers often glance there using only eye movement. The authors introduced an "owlness" metric to characterize individual gaze strategies, defined as the ratio of head movement distance to the sum of head and eye movement distances. Drivers classified as "owls" (high head movement) showed little to no accuracy improvement from adding eye pose, whereas "lizards" (low head movement, high eye movement) saw significant gains. This metric effectively explained the inter-person variation in classification performance, with "lizard" strategies benefiting most from eye pose data. The significance of this work lies in its characterization of how gaze strategies influence the utility of eye pose information in driver monitoring systems. By identifying that eye pose is most valuable for drivers who rely on eye movements rather than head turns, the study provides insights for designing more robust and personalized gaze tracking algorithms. This understanding can enhance the effectiveness of distraction detection and alerting systems, particularly for drivers who maintain a fixed head position while shifting visual attention.

Key finding

Adding eye pose on top of head pose increases six-region gaze classification accuracy from 89.2% to 94.6% on average (a 5.4 percentage point gain) at ~1.3 decisions/second; the gain is concentrated in 'lizard' drivers who keep the head still while 'owl' drivers see little or no improvement, motivating an 'owlness' metric to explain inter-individual differences.

Methodology

on_road

Sample size: 40 drivers; ~1,351,864 annotated frames across 6 gaze regions

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_cs.HC 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.

• gaze based attention detection
• distraction detection algorithms
• drowsiness detection algorithms
• eye movements scanning
• looked but failed to see
• attention allocation

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: behavioral performance data
• Methodological Resource: tool software, measurement protocol