Hands on the wheel: A Dataset for Driver Hand Detection and Tracking

Borghi, Guido; Frigieri, Elia; Vezzani, Roberto; Cucchiara, Rita · 2018 · Crossref

DOI: 10.1109/fg.2018.00090

archive: archived pipeline: cataloged verified

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

Summary

This paper addresses the critical need for reliable driver hand detection and tracking in automotive contexts to monitor attention levels, predict maneuvers, and enable new Human-Car Interaction systems. The authors identify that existing methods often struggle with varying illumination, occlusions, and limited viewpoints. To overcome these challenges, they introduce the Turms dataset and a novel real-time detection method. The motivation stems from the increasing prevalence of in-car cameras and the safety imperative to detect manual distractions, such as hands off the wheel, which contribute significantly to fatal accidents. The study presents the Turms dataset, comprising over 14,000 infrared stereo frames captured using a Leap Motion device placed at the back of the steering wheel. This innovative rear viewpoint minimizes body occlusions and allows for a wide field of view. The dataset features seven subjects performing various driving actions, with manual annotations for hand positions and steering wheel contours. The proposed detection method operates in three stages: image pre-processing, homography transformation, and image unrolling. First, the system rectifies the fish-eye images and fits an ellipse to the steering wheel’s external border. Second, it applies a homography transformation to map the elliptical wheel into a circular space, determined via an iterative algorithm that minimizes radius and center discrepancies. Third, the circular image is unrolled into a linear space, simplifying the detection task. In the final stage, hand presence is detected by analyzing the intensity histogram of the unrolled image. The algorithm identifies local maxima on either side of a central peak (caused by the device’s infrared LEDs), which correspond to the driver’s hands. The system determines hand location based on the angle derived from the histogram peaks. Experimental results demonstrate that the method achieves over 95% accuracy in predicting the presence of both hands on the wheel. The system operates in real time at more than 30 frames per second on standard hardware, with low memory requirements. The authors note that while the method is robust for the specific rear-wheel viewpoint, it cannot detect hands in the central histogram region due to interference from the infrared light source. The significance of this work lies in providing a specialized dataset and a computationally efficient algorithm tailored for the unique constraints of in-vehicle monitoring. By leveraging infrared stereo imaging from a rear perspective, the approach offers a non-invasive solution that functions reliably under varying light conditions. The authors conclude that this method serves as a foundational step for developing advanced driver assistance systems and gesture-based interfaces, allowing drivers to interact with vehicle controls without removing their hands from the steering wheel. Future work aims to incorporate tracking algorithms for temporal coherence and expand the dataset to include diverse skin tones and glove usage.

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.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

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

• distraction detection algorithms
• manual
• hands on hands off engagement

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).

• Methodological Resource: tool software