VISUAL ATTENTION BEFORE CROSSING01 — PROJECT OVERVIEW

Before the Step

Visual Attention and Safety Mechanisms Before Pedestrian Crossing Initiation

How does visual attention before crossing initiation relate to pedestrian safety?

Using VR experiments and eye tracking with 97 participants, this study examines visual attention before pedestrian crossing initiation.

Visual reorganization begins approximately 6.25 seconds before the first step. Larger crossing safety margins are associated with greater relative attention to vehicles, less attention to roadside objects and the crosswalk, and less frequent visual switching.

RESEARCH BY

Prof. Si Qiao · Mr Yiwei Mo

CITE THIS WEBSITE

Si Qiao & Yiwei Mo. (2026). Before the Step: Visual Attention and Safety Mechanisms Before Pedestrian Crossing Initiation. https://si-lab.city/data/before-the-step/

Illustration of a pedestrian observing vehicles and the crossing environment before stepping onto a zebra crossing
Before the first step: observing traffic, rules, the path, and the surrounding environment.Conceptual illustration
VR Experiment · Eye Tracking · Pedestrian Behavior · Crossing Safety
01.1

Key Finding

Crossing safety depends not only on when pedestrians step forward, but also on how they organize visual attention beforehand.

The results indicate that attentional resource allocation, the timing of vehicle information entering visual processing, and the organization of visual search before movement initiation are associated with subsequent crossing safety margins. When these visual processes are considered together, attentional allocation and overall visual switching patterns show more stable independent associations.

01 Attentional Allocation
02 Timing of Vehicle Information
03 Visual Search Organization
02 — RESEARCH CONTEXTWHY BEFORE THE STEP?

Why before the step?

Understanding pedestrian safety begins before movement begins.

Public context data, separate from this study. The annual death estimate (1.16 million) follows the WHO road safety overview. Vulnerable road users (>50%; reported as 53%) and pedestrians (23%) follow the WHO Global status report on road safety 2023, based on 2021 data. These sources use different reporting periods; the 2023 report estimates 1.19 million total deaths in 2021.

FROM REAL-WORLD RISK TO THE RESEARCH QUESTION

THE REAL-WORLD CHALLENGE

The risk begins
before the crossing.

TRADITIONAL FOCUS

CollisionsWhether to crossWhen to stepSafety margins

THIS STUDY ASKS ONE STEP EARLIER

What happens visually
before the first step?

VehiclesRulesPathEnvironment

How does visual attention gradually become organized?

03 — EXPERIMENTAL STUDYVIRTUAL REALITY · EYE TRACKING

VR-based Pedestrian
Crossing Experiment

This controlled VR pedestrian crossing study collected eye tracking, traffic states, and crossing behavior under different vehicle speeds and initial distances to examine how visual processes before movement initiation relate to subsequent crossing safety margins.

01

PARTICIPANTS

The study sample

97Participants
550Valid Trials
23.96 ± 5.17Mean Age ± SDAge range: 18–58 years

All participants completed the VR crossing task with eye tracking.

Participant wearing a VIVE headset
Participant view 01 · wearing the headset
Participant adjusting a VIVE headset
Participant view 02 · adjusting the headset
02

APPARATUS & VIRTUAL ENVIRONMENT

The equipment and the street participants experienced

A. Apparatus

VIVE headset, controllers, tracking base stations and cables
HTC VIVE equipment · walking platform not pictured
HTC VIVE Pro Eye

VR display & eye tracking

Walking platform

Physical walking interaction

Unity

Virtual environment & traffic control

B. Virtual Environment

Unity editor showing the urban crossing scene and traffic controller
Unity scene construction
Overhead view of the urban junction and crossing paths
Overhead view · crossing area and vehicle paths

The urban zebra-crossing junction was built in Unity. Vehicle paths, vehicle spawning, and traffic conditions were controlled programmatically.

Participant View

First-person VR view of vehicles, signal equipment, sidewalk and roadside buildings
Participant view 01 · street-level visual information
First-person crossing view annotated with Traffic, Rule, Path, and Context areas of interest
Participant view 02 · four AOI information sources
01 Traffic Moving vehicles02 Rule Signals & traffic rules03 Path Crossing path04 Context Roadside environment
03

EXPERIMENTAL DESIGN

Traffic Scenario → Visual Processing → Crossing Safety

Traffic Scenario→Visual Processing→Crossing Safety
01

Traffic Scenario

VEHICLE SPEED25 / 35 / 45 km/h
×
INITIAL DISTANCE30 / 50 m
↓
6TRAFFIC CONDITIONS
02

Visual Processing

INFORMATION SOURCES · 4 AOI TYPES

RuleSignals & traffic rules
TrafficMoving vehicles
Visual attention
ContextRoadside environment
PathCrossing path
ANALYSIS · HOW ATTENTION IS ORGANIZED
Attention
AllocationWhere & how much
Event
TimingWhen to look
Transition
OrganizationHow gaze switches
03

Crossing Safety

Dmin↔

Crossing Safety Margin

Minimum pedestrian–vehicle distance during crossing

↔
Small marginSmaller spatial buffer
↔
Large marginLarger spatial buffer
04

DATA COLLECTION

Four complementary measurement streams

Eye Tracking

Gaze & AOI

  • Gaze direction
  • AOI state
  • Fixation sequence

Traffic

Vehicle State

  • Speed
  • Initial distance
  • Vehicle position

Behavior

Pedestrian Movement

  • Crossing initiation (M₀)
  • Position
  • Walking trajectory

Safety Outcome

Crossing Safety Margin

  • Minimum pedestrian–vehicle distance
  • Dmin
04 — VISUAL REORGANIZATION BEFORE CROSSING97 PARTICIPANTS · 550 TRIALS

Visual Reorganization
Before Crossing

Aligned to crossing initiation (M0), continuous AOI data reveal when visual attention departs from its early baseline and define the window for subsequent analysis.

01

CONTINUOUS VISUAL SEQUENCE

One recorded trial · 50 ms resolution

Actual AOI sequence from one full pre-action trial, including unresolved states
A single recorded trial, not a group average. Gray marks unresolved AOI states.Chart data ↓
02

VISUAL REORGANIZATION TRAJECTORY

Fitted change in Traffic–Context contrast from the early baseline

Real fitted Traffic–Context contrast change and 95 percent simultaneous confidence band, with onset at minus 6.25 seconds
Early baseline: −10 to −8 s. Estimates and confidence band from the supplied analysis output.Chart data ↓
95% simultaneous confidence bandSustained change ≥ 1 s
−6.25 sVisual reorganization onset

VISUAL CHANGE BEFORE ACTION

The contrast begins a sustained departure from the early baseline at −6.25 s. The detected departure lasts to approximately −0.15 s; the analysis window extends to M0.

Earlier observationVisual reorganizationAction
−10 s−6.25 sM0 · 0 s
Visual mechanism analysis window · 6.25 s

Attention allocation, key event timing, and visual state transition measures are extracted within this unified window.

01Attention Allocation02Event Timing03Transition Organization
05 — HOW DOES VISION REORGANIZE?WHERE · WHEN · HOW

How Does Vision
Reorganize?

Before the first step, visual attention reorganizes through where attention is allocated, when traffic information is first attended to, and how gaze switches between information sources. All three dimensions are associated with crossing safety margins.

01 · WHERE

Attention Allocation

Where is visual attention allocated?

TrafficRulePathContext

More Traffic attention relative to Rule→ Larger safety margin

More crosswalk attention→ Smaller safety margin

Actual attention allocation model coefficients with 95 percent confidence intervals; Traffic to Rule attention ratio positive, crosswalk attention negative, roadside object estimate weaker after correction

Points: change per 1 SD. Lines: 95% CI.

Largest explanatory gain among the three dimensions.

Model R² 0.249ΔR² +0.088
02 · WHEN

Event Timing

When is traffic information first attended to?

Longer First Traffic Fixation Delay→ Smaller safety margin

Delay from the analysis window start (−6.25 s) to the first Traffic attention event.

Earlier attention↓Larger margin
Later attention↓Smaller margin

Other event timings did not show clear associations after multiple-comparison correction.

Model R² 0.229ΔR² +0.068
03 · HOW

Transition Organization

How is visual search organized?

Frequent Switching

Traffic→Rule→Path→Context→Traffic

Higher Overall AOI Switching Rate→ Smaller safety margin

Sustained Monitoring

TrafficTraffic

More Sustained Traffic Segments→ Larger safety margin

Sustained segments are continuous Traffic monitoring for at least 200 ms.

Model R² 0.225ΔR² +0.064

Patterns associated with a larger crossing safety margin

WHERE · ATTENTION ALLOCATION

Greater attention to Traffic relative to Rule
Less attention to the crosswalk

WHEN · EVENT TIMING

Earlier first attention to Traffic

HOW · TRANSITION ORGANIZATION

Less frequent AOI switching
More sustained Traffic monitoring

Selected result data ↓
06 — CORE FINDINGJOINT-MODEL SYNTHESIS

Core Finding

Crossing Safety Margin
+ Traffic / Rule Attention− Roadside Attention− Crosswalk Attention− AOI Switching

Traffic / Rule ↑ · Roadside ↓ · Crosswalk ↓ · AOI Switching ↓

Larger crossing safety margins are associated with a more traffic-focused, less dispersed, and less frequently switching visual pattern before action.

Key Visual Indicators

01Traffic / Rule
Attention Ratio
02Roadside-object
Attention
03Crosswalk
Attention
04Overall AOI
Switching Rate
07 — ADDITIONAL SUGGESTIONSDESIGN IMPLICATIONS

From Visual Mechanism
to Safer Crossing Design

If safer crossing is associated with more traffic-focused and less fragmented visual processing, street design should help pedestrians identify vehicle information earlier, reduce unnecessary visual competition, and maintain a stable view of approaching traffic.

CORE FINDING

Traffic-focused + Less fragmented vision

01

Make Traffic Information
Easier to See

Traffic / Rule Attention ↑

  • Keep clear sightlines toward approaching vehicles.
  • Reduce vehicle occlusion from railings, greenery, and street facilities.
  • Make vehicle movement visible from key observation positions.

Support earlier and clearer monitoring of approaching vehicles.

Clear sightlines and reduced obstruction help make approaching vehicle movement easier to see.
Clear sightlines and reduced obstruction help make approaching vehicle movement easier to see.Conceptual design illustration · Click to enlarge
02

Reduce Competing
Visual Information

Roadside Attention ↓
Crosswalk Attention ↓

  • Limit excessive advertising, complex street furniture, and visually distracting elements in the decision zone.
  • Keep path guidance simple and easy to read.
  • Give critical traffic information a clear visual priority.

Keep the crossing decision zone visually simple.

Simpler roadside information and path guidance give critical traffic information visual priority.
Simpler roadside information and path guidance give critical traffic information visual priority.Conceptual design illustration · Click to enlarge
03

Support Stable
Visual Search

AOI Switching ↓

  • Keep signals, approaching traffic, and crossing entrances within a coherent, easy-to-scan field of view.
  • Reduce layouts that require repeated searching for critical information.
  • Avoid scattering key traffic information across multiple directions.

Reduce unnecessary visual switching between critical information sources.

A coherent field of view keeps key information together and reduces repeated searching.
A coherent field of view keeps key information together and reduces repeated searching.Conceptual design illustration · Click to enlarge