RightGazeUrban-2022: A Spatiotemporal Gaze Trajectory Dataset of Drivers in Urban Right-Turning Cyclist Conflict Scenarios

This dataset contains spatiotemporal gaze and vehicle trajectory data from a controlled test track experiment investigating how drivers scan for cyclists when making right turns at urban intersections. It captures the gaze behavior, vehicle kinematics, and driver demographics of n = 34 participants who performed 539 standardized right-turn maneuvers on a closed outdoor test track under conditions where a cyclist could approach from behind on the right side — a common and dangerous conflict type in urban traffic. Only right-turning trials are included in this release; the straight-going trials that were part of the original study are not included. The dataset was collected in 2022 and is described in full detail in the associated scientific article (see Related Publication below). Related Publication (Open Access) Denk, F., Fröhling, F., Brunner, P., Huber, W., Margreiter, M., Bogenberger, K., & Kates, R. (2025). Influence of gaze strategies and cognitive load on safeguarding performance of motorists in right-turning scenarios involving potential conflicts with vulnerable road users. Transportation Research Part F: Traffic Psychology and Behaviour, 109, 32–49. https://doi.org/10.1016/j.trf.2024.11.012 The article provides the full study design, methodology, statistical analysis, and scientific interpretation of the data contained in this repository. Study Design Participants drove a driving school Seat Leon on a closed outdoor test track. The road geometry is specified in the included OpenDRIVE file (geometry.xodr): a right turn with a curve radius of 16.75 m and a cycle lane offset of 7 m, with no occlusion between the driver and the potential conflict zone. During the right-turning maneuver, a cyclist could approach from the rear-right — a direction that requires active scanning beyond the typical forward field of view. A human actor cyclist was incorporated into the study for situational awareness; however, due to ethical constraints, the cyclist was only present during straight-going maneuvers (hiding in a parked van otherwise) and was never on a direct collision course with participants. The study was structured as a within-subject design with three independent factors: Maneuver type: right-turning and straight-going Target velocity: 10, 20, 30, 40, and 50 km/h Cognitive load: auditory n-back task at level 1 (easier) and level 2 (harder), with cues every 2.25 s Each participant completed 22 trials in total (16 right-turning, 6 straight-going). Participants were randomly assigned to one of four randomized trial sequences to control for order effects. Straight-going trials were included to conceal the true focus of the study (right-turn safeguarding) from participants. Gaze data were manually labeled from on-board video footage into five Areas of Interest (AOIs): AOI Definition LF Looking front RVM Rear view mirror SVM_R Right side view mirror WG_R Right side window SG_R Shoulder glance behind the car’s B-pillar The polygon geometry underlying these AOIs is provided in aoi_config.json (see Dataset Contents below). It is required by the EQUGAT gaze trajectory analysis algorithm which can also be used to visualize participants’ gaze trajectories using the companion repository GazeTrajectoryAnalysis. Vehicle position (ADMA with IMU and DGNSS), speed, and gaze data were synchronized and sampled at 100 Hz. Participants n = 34 participants were recruited and completed the experiment (20 male, 14 female) Age range: 20–66 years (M = 37, SD = 13) Driving experience: M = 19.7 years (SD = 13.3; range: 2–49 years) Inclusion criterion: valid driver’s license; exclusions for severe visual impairments or physical limitations preventing shoulder glances Ethical approval: GEHBa — Gemeinsame Ethikkommission der Hochschulen Bayerns Trial statistics 770 test drives were conducted in total. One participant had missing GNSS data for all drives (excluded entirely, see Participants above); five further drives from other participants had incomplete GNSS data, leaving 743 valid drives, of which 539 were right-turning maneuvers. Only these 539 right-turning trials are included in this release — the remaining straight-going trials, which were included in the original study for concealment purposes only (see Study Design above), are not published. Dataset Contents File Description gazes_spatiotemporal.feather Spatiotemporal gaze and trajectory data (primary dataset) demographics_and_questionnaire_data.xlsx Selected driver demographics and questionnaire responses scenario_data.xlsx Per-trial scenario parameters (velocity, n-back level, sequence assignment) and trial-level self-report/performance measures geometry.xodr OpenDRIVE road network file defining the test track scenario geometry aoi_config.json Polygon geometry of the five gaze AOIs, as used by the EQUGAT gaze trajectory analysis algorithm (see GazeTrajectoryAnalysis) Variable Descriptions gazes_spatiotemporal.feather This file contains one row per time step per trial. It is stored in the Apache Arrow Feather format (version 2) for efficient columnar access. Data are sampled at 100 Hz. Column Type Description trajectory_id int Anonymized unique identifier for a single trial (one driver completing the right-turn maneuver once) driver_id int Anonymized unique identifier for a participant (driver) t float Time within the trial (seconds) s_driver float Longitudinal position of the vehicle along the road centerline (meters), measured as distance from the start of the scenario segment (x = 0.0 m) v float Vehicle speed (m/s) gaze_state categorical Classified gaze AOI at this time step: one of LF, RVM, SVM_R, WG_R, SG_R (see AOI table above) gaze_id int Anonymized identifier for the individual gaze fixation/event; value 0 indicates no classified gaze object (transition or undefined state) demographics_and_questionnaire_data.xlsx One row per participant. Column Type Description driver_id int Anonymized unique participant identifier, consistent with gazes_spatiotemporal.feather age_group categorical Age group of the participant (bins: 20–29, 30–39, 40–49, 50–59, 60–69) gender categorical Self-reported gender (male / female) CarUrbanUsage ordered categorical Self-reported frequency of urban car driving (never / seldom / sometimes / often / always) CyclingUsage ordered categorical Self-reported frequency of cycling (never / seldom / sometimes / often / always) Expectancy ordered categorical Agreement with: “During the experiment I expected crossing cyclists from behind” — 5-point Likert scale (disagree / somewhat disagree / undecided / somewhat agree / agree) Responsibility ordered categorical Agreement with: “In the experiment I felt the same responsibility regarding endangering other road users as in real traffic” — 5-point Likert scale (disagree / somewhat disagree / undecided / somewhat agree / agree) scenario_data.xlsx One row per trial (trajectory). Column Type Description trajectory_id int Anonymized unique trial identifier, consistent with gazes_spatiotemporal.feather driver_id int Anonymized unique participant identifier, consistent with gazes_spatiotemporal.feather and demographics_and_questionnaire_data.xlsx randomized_sequence_number int Which of the four randomized trial sequences (1–4) the participant was assigned to (see Study Design above) maneuver categorical Maneuver type for this trial; always right-turning in this release (see Trial statistics above) target_approach_velocity int Target approach velocity for this trial (km/h): one of 10, 20, 30, 40, 50 nback int Auditory n-back task level for this trial: 1 (easier) or 2 (harder) response_attention_slider float Post-scenario self-reported attention rating (1–5 slider scale) nback_performance float Proportion of correct responses on the n-back task during this trial (0–1) geometry.xodr An OpenDRIVE 1.x road network file specifying the geometry of the test track intersection, including lane widths, road curvature (R = 16.75 m), and the lateral offset of the cycle lane (7 m from the car lane center). aoi_config.json Defines the polygon geometry (vehicle-relative coordinates) of the five gaze AOIs listed above, along with a display color and alpha value for each. This file is required by the EQUGAT gaze-classification algorithm used to produce gaze_state in gazes_spatiotemporal.feather, and can be used together with the companion repository GazeTrajectoryAnalysis to visualize participants’ gaze trajectories against the AOI polygons. Loading the Data Python import json import pandas as pd # Gaze data df_gazes = pd.read_feather("gazes_spatiotemporal.feather") # Demographics df_demo = pd.read_excel("demographics_and_questionnaire_data.xlsx") # Scenario data df_scenarios = pd.read_excel("scenario_data.xlsx") # AOI polygon definitions with open("aoi_config.json") as f: aoi_config = json.load(f) Requires pandas >= 1.3 and pyarrow for Feather support: pip install pandas pyarrow openpyxl R library(arrow) library(readxl) library(jsonlite) df_gazes <- read_feather("gazes_spatiotemporal.feather") df_demo <- read_excel("demographics_and_questionnaire_data.xlsx") df_scenarios <- read_excel("scenario_data.xlsx") aoi_config <- fromJSON("aoi_config.json") Requires the arrow, readxl, and jsonlite packages. Anonymization The dataset has been anonymized to comply with data protection requirements: Age has been binned into 10-year age groups; exact ages are not released. Driver IDs (driver_id) have been replaced with random integers drawn without replacement from the range 1–100,000. The original pseudonymous laboratory IDs are not included. The mapping is consistent across gazes_spatiotemporal.feather, demographics_and_questionnaire_data.xlsx, and scenario_data.xlsx. Trajectory IDs (trajectory_id) have been replaced with random integers in the same range. The mapping is consistent across gazes_spatiotemporal.feather and scenario_data.xlsx. Gaze IDs (gaze_id) have been replaced with random integers, with the exception of gaze_id = 0 (no gaze object), which is preserved as-is. Row order in the gaze data has been sorted by trajectory_id and t, and rows in the demographics and scenario data have been sorted by ID, to avoid any implicit ordering that could allow re-identification of participants. Funding This work was supported by the Technische Hochschule Ingolstadt. Citation If you use this dataset, please cite both the dataset and the associated article: Dataset: Denk, F., Fröhling, F., Brunner, P., Huber, W., Margreiter, M., Bogenberger, K., & Kates, R. (2025). RightGazeUrban-2022: A Spatiotemporal Gaze Trajectory Dataset of Drivers in Right-Turning Cyclist-Conflict Scenarios [Data set]. Zenodo. https://doi.org/10.5281/zenodo.20717040 Article: Denk, F., Fröhling, F., Brunner, P., Huber, W., Margreiter, M., Bogenberger, K., & Kates, R. (2025). Influence of gaze strategies and cognitive load on safeguarding performance of motorists in right-turning scenarios involving potential conflicts with vulnerable road users. Transportation Research Part F: Traffic Psychology and Behaviour, 109, 32–49. https://doi.org/10.1016/j.trf.2024.11.012 Contact For questions about the dataset please reach out via Zenodo or GitHub!