Enhanced Light Crosswalk Illumination

The System

SafeWalk™ prototype with seven adjustable spectra

Purpose-built LED crosswalk illuminator (TAPCO SafeWalk 2.0), configured for field research.

SafeWalk crosswalk illuminators mounted on two poles lighting a pedestrian at a midblock crosswalk at night — SafeWalk™ crosswalk illuminator: dual-pole layout lighting the pedestrian, not just the pavement

≥ 20 lux vertical at pedestrian plane 12–15 ft mounting Adjustable beam aiming 25° × 24° beam · ≈820 cd/klm IP66 · −40°F to 176°F 120 mph wind rating ≈2 W per LED · 12 V DC Solar + battery compatible

Prototype spectral configurations assembled for testing — Prototype modules assembled for testing: 4000 K · 5000 K · 6500 K · RGB blue-light

Configuration	Type	Rationale
4000 K	Neutral white	FHWA / agency reference
5000 K	Cool white	One step above standard
6500 K	Daylight white	Upper white-light boundary
Blue Baseline	Blue-shifted LED	Reference blue configuration
Ice Blue	Blue-shifted LED	Blue + white mix, highest luminance
Sky Blue	Blue-shifted LED	Mid-intensity mix
Pure Royal Blue	Saturated blue	Maximum blue saturation

Seven spectral configurations: the agency-recommended white range plus the blue-shifted range never before field-tested in Wisconsin.

Field Testing

281 nighttime rounds at two Wisconsin sites

An instrumented probe vehicle approached the crosswalk while a mannequin on a remote-controlled electric skateboard crossed, creating repeatable pedestrian events on real roads, on real nights.

Madison site map and baseline lux measurement

Madison · 104 rounds

Williamson St & S Dickinson St

Wider 4-lane urban roadway · existing RRFB + push buttons · visually complex background.

Baseline ≈ 0.5 lux Heights: 10 / 12.5 / 15 ft RRFB present

Milwaukee site map with U-turn test route

Milwaukee · 177 rounds

N Maryland Ave · UWM campus

Two-lane campus roadway · no RRFB or push button · ambient street lighting.

Baseline ≈ 13 lux Heights: 7.5 / 10 / 12.5 ft Lateral: 0 / 9.5 / 16.5 ft

What we varied

Factor	Levels
Lighting spectrum	4000 K · 5000 K · 6500 K · 4 blue configs
Mounting height	3 heights per site
Lateral distance	0 / 9.5 / 16.5 ft (Milwaukee)
Pedestrian target	Adult & child mannequins
Clothing	Dark blue & white
Obstruction	Clear · small vehicle · van parked
Weather / ambient	Clear night · light rain · ± streetlight

Instrumentation

Field hardware: mannequins on skateboards, cameras, GPS, brake logger, lux meter, test vehicles — Field hardware: mannequin skateboards · 4K dual cameras · brake foot switch · lux meter · GPS · two sedans + one van

125 Hz GPS Roadway camera Driver IR camera Brake logger Lux meter Remote skateboard control

Drivers were never told when or where the mannequin would cross.

Measured illuminance at the crosswalk (Madison)

≈ 0.5 luxBaseline · no enhanced lighting
≈ 21 luxEnhanced crosswalk lighting
≈ 40 luxEnhanced + floodlight (streetlight sim.)

Lux meter readings: 0.5, 21, and 40 lux — Lux meter readings: (a) 0.5 lux baseline · (b) 21 lux enhanced · (c) 40 lux with floodlight

Nighttime field testing setup at Madison and Milwaukee — Nighttime testing: (a) Madison, wet pavement · (b) Milwaukee, blue-lit crosswalk

Field effort under harsh and tough conditions

All 281 rounds ran at night, including light rain, wet pavement, and late-season cold.

Instrumented test vehicle on wet night street

Wide view of the wet Madison test street at night

How We Measured

Deep-learning eye tracking of real drivers

A CNN pipeline found the exact frame each driver first recognized the pedestrian, then synchronized it with GPS, speed, and brake records.

Collect

GPS · roadway video · driver IR video · brake logs, all synchronized.

Synchronize

3.5 s event clips centered on each brake press.

Detect

CNN face landmarks → gaze vector → first recognition frame.

Measure

RT = brake − eye onset · DD = distance at onset.

CNN face landmark localization — (a) CNN face landmarks

Pupil center and gaze vector estimation — (b) Pupil center & gaze vector

Frame-level gaze timeline with eye onset — (c) Gaze timeline → eye onset → RT

Reaction Time

First recognition → brake press. Shorter is better.

Detection Distance

Vehicle-to-crosswalk distance at first recognition. Longer is better.

GPS trajectory mapping for the Madison experiment — 125 Hz GPS trajectory mapping: Madison approach path verification

Quality control

Every eye onset was verified against the driver video, the roadway video, and the brake-event timestamp before analysis.

Face detection 95.6% combined 93.8% Milwaukee 98.6% Madison 30 fps IR video · ≈33 ms resolution Blinks, saccades & head pose filtered from noise

Only trials passing synchronization, video, GPS, and brake-match checks entered the final analysis.

Watch

See the system in action

Live nighttime demonstrations at both sites, and the eye-tracking pipeline running on driver video.

Enhanced lighting demonstrations at the Madison and Milwaukee sites — Live demonstrations at Madison (a, b) and Milwaukee (c–e)

Live demonstrations for the WisDOT POC

Held at both study sites under real nighttime conditions. The committee observed illumination coverage, vehicle approach, and mannequin crossings first-hand.

Madison: wider urban roadway · RRFB · wet pavement Milwaukee: two-lane campus roadway · pedestrian activity Strong interest in the blue-light option at SE WI Transportation Symposium 2025

▶ Add YouTube ID for the Milwaukee demo
(edit VIDEO_IDS in index.html)

Milwaukee Demonstration

N Maryland Ave · two-lane campus roadway

▶ Add YouTube ID for the Madison demo
(edit VIDEO_IDS in index.html)

Madison Demonstration

Williamson St & S Dickinson St · RRFB site

▶ Add YouTube ID for the eye-tracking demo
(edit VIDEO_IDS in index.html)

Eye-Tracking Pipeline

Driver-facing infrared video · onset detection

Demonstrated live to the WisDOT Project Oversight Committee at both sites, and at the SE Wisconsin Transportation Symposium 2025.

Results

Blue-spectrum lighting won at both sites

Baseline vs. enhanced lighting: measured driver reaction time and detection distance.

Madison · Blue Baseline

Baseline · RT 0.71 s212 ft

Enhanced · RT 0.67 s275 ft

+63 ft earlier detection · faster reaction

Milwaukee · Ice Blue

Baseline · RT 0.92 s242 ft

Enhanced · RT 0.69 s261 ft

+19 ft earlier detection · RT −0.23 s

Overall: blue-spectrum enhanced lighting let drivers detect the pedestrian ~40 ft sooner than baseline and ~26 ft sooner than conventional 4000 K white, with comparable or faster reaction times.

Spectrum comparison

Spectrum	RT (s)	DD (ft)
Baseline (no enhanced)	0.92	242.0
4000 K	0.72	237.2
5000 K	0.82	239.4
6500 K	0.75	244.6
Blue Baseline	0.73	256.4
Ice Blue ★	0.69	260.8
Pure Royal Blue	0.83	239.9
Sky Blue	1.13	254.8

Ice Blue strongest overall: one of the fastest reaction times with a practical detection distance. Blue Baseline beat 4000 K by +19.2 ft DD at similar RT.

Milwaukee spectrum performance comparison — Milwaukee spectrum performance: RT & DD

Milwaukee lateral distance comparison — Lateral placement: best: 9.5 ft

Milwaukee mounting height comparison — Mounting height: best: 12.5 ft

Recommended Milwaukee configuration

Spectrum: Ice Blue Mounting height: 12.5 ft Lateral distance: 9.5 ft Spill control: none

Strongest balance of RT, DD, sample size, and installation practicality: the two-lane design basis.

Mannequin type

Adult vs. child differences were small relative to overlapping error bars.

Spill control

Crosswalk-only lighting showed the longest DD; limited trials, reference only.

Obstruction

A parked vehicle sharply reduced detection distance; the main effect is later recognition.

Milwaukee combined configuration comparison — Combined comparison: spectrum · lateral distance · mounting height

Spectrum comparison

Spectrum	RT (s)	DD (ft)
Baseline (no enhanced)	0.71	212.0
4000 K	0.66	241.8
5000 K	0.77	251.3
6500 K	0.94	240.0
Blue Baseline ★	0.67	275.0
Ice Blue	1.20	221.3
Pure Royal Blue	0.62	209.1
Sky Blue	1.13	215.9

Blue Baseline strongest balanced result: +33.2 ft DD over 4000 K at unchanged RT, and +63 ft over baseline.

Madison spectrum performance comparison — Madison spectrum performance: RT & DD

Madison mounting height comparison — Mounting height: best: 15.0 ft

Madison combined spectrum and height comparison — Combined spectrum + height comparison

Recommended Madison configuration

Spectrum: Blue Baseline Mounting height: 15.0 ft Lateral: site-specific Spill control: none

Lateral distance was not independently varied at Madison; placement was set through site review and roadway-width scaling.

Mannequin type

The child mannequin had the highest RT and shortest DD; drivers recognized the smaller target later.

Spill control

50% spill control looked favorable but with very few trials; reference only.

Obstruction

Results varied by condition; drivers may have approached the larger van more cautiously.

Cross-site synthesis

Site-specific spectrum decision rankings — Spectrum ranking: (a) Milwaukee: Ice Blue · (b) Madison: Blue Baseline

The winner depends on the road.

Ice Blue best for the two-lane residential (Milwaukee) context. Blue Baseline best for the wider urban (Madison) context, and carried forward as the generalized recommendation.

Clothing condition effects by lighting spectrum — Clothing effects: Ice Blue best at Milwaukee for both dark & white clothing; Blue Baseline best at Madison

Eye-tracking physiological metrics (N = 163)

Metric	Baseline	Enhanced	p	Meaning
Pupil diameter	5.31 mm	4.87 mm	<0.001	Pupil constriction under enhanced light
Contrast ratio	0.72	0.88	<0.001	Higher pedestrian–background contrast ✓
Glare index	0.18	0.31	<0.001	Increased: monitor glare in design ⚠
Visual comfort score	3.82	3.41	0.002	Slight comfort reduction
Blink rate	0.31 /s	0.22 /s	<0.001	Reduced blink rate
Pupil asymmetry	0.08	0.13	<0.001	Greater asymmetry; possible glare indicator ⚠
Eye aspect ratio	0.284	0.298	0.041	Slightly higher under enhanced lighting
Saccade velocity	142.3°/s	168.7°/s	<0.001	Faster visual scanning ✓

Contrast and saccade gains support the visibility benefit; glare metrics justify glare review in final design.

Final recommended lighting specifications

Application context	Spectrum	Height	Lateral	Power
2-lane residential (Milwaukee-type)	Ice Blue	12.5 ft	9.5 ft	Solar + battery
Wider urban (Madison-type)	Blue Baseline	15.0 ft	Site-specific	Solar + battery
Generalized guidance	Blue Baseline	Width-scaled	Width-scaled	Solar + battery

Design illuminance ≥ 20 lux vertical at the pedestrian plane (≈ 21 lux measured in the field) · no spill control required, but glare reviewed site-specifically.

Design Guidelines

From two sites to any Wisconsin crosswalk

A planning-level framework, not a fixed standard. Every site still gets engineering review and nighttime verification.

Describe context

Width · lanes · speed · ambient lighting

Apply Blue Baseline

Spectrum from field results

Scale geometry

Height & offset from roadway width

Predict RT & DD

Blue Baseline OLS model

Screen & review

DD vs. SSD · glare · ADA

Roadway geometry scaling for mounting height and lateral placement — Height (a), lateral placement (b) & illumination angle (c) vs. road width. **Example:** 44 ft four-lane roadway → ≈ 16.0 ft height, ≈ 11.0 ft lateral, then verified against pole feasibility, glare, and coverage

Blue Baseline OLS models

Variable	RT	DD
Mounting height	0.038 *	1.452
Lateral distance	−0.004	−1.271
Road width	−0.001	1.395
Vehicle speed	0.012	7.244 **
Child vs. adult	0.106	−34.568
High vs. medium ambient	0.065 **	3.457
R²	0.208	0.451

* p < 0.1 · ** p < 0.05. The DD model (R² = 0.451) is the useful screen; predicted DD compares directly with SSD. Exploratory planning tool only.

Stopping sight distance adequacy across roadway types — SSD adequacy: green: predicted DD > required SSD · red: needs design revision

Stopping sight distance screen

Predicted detection distance must exceed AASHTO-required SSD (2.5 s perception-reaction, f = 0.35, level grade).

Pass School-zone, residential, downtown, urban, most suburban

Marginal High-speed 4-lane divided: verify before advancing

Fail Rural 4-lane · high-speed 6-lane divided: revise design first

Required checks before installation

Photometric coverageVertical illumination Glare reviewBeam direction ADA clearanceClear zone UtilitiesSolar access Battery maintenanceElectrical code Median pole reviewWinter durability

For WisDOT Review

Implementation recommendations

Design element	Recommended direction
Primary spectrum	Blue Baseline (blue-shifted LED; controller RGB R 129.0 · G 420.75 · B 741 on the 0–750 scale) · Ice Blue (R 80 · G 520 · B 750) for narrow two-lane residential contexts
Vertical illuminance	≥ 20 lux at the pedestrian plane (≈ 21 lux measured in the field)
Mounting height	Scale with roadway width (10.6 ft @ 22 ft road → 19.2 ft @ 72 ft road) · verify photometrically
Lateral placement	Scale with roadway width (5.6–13.7 ft) · adjust for poles, glare, coverage
Power source	Solar + battery storage, subject to site feasibility
Safety screening	Predicted detection distance must exceed required SSD for the approach speed

1 · Lighting works

Blue-spectrum lighting reduced RT and increased DD vs. baseline at both sites.

2 · Spectrum alone isn't enough

It must combine with illuminance, contrast, coverage, glare control, and geometry.

3 · Optimal spectrum is site-dependent

Ice Blue for two-lane; Blue Baseline for wider roads and generalized use.

4 · Geometry drives transfer

Scale height and placement with roadway width; never copy between sites.

5 · SSD is the safety screen

Predicted DD must exceed required SSD, or the site needs additional review.

Next steps

Pilot deployments · larger datasets · adaptive dimming · winter validation · glare monitoring · statewide guidance updates.

Comments

Questions, corrections, and suggestions

Reviewing this study? Send a comment to the research team at UW–Milwaukee. Technical questions, data requests, implementation suggestions, and corrections are all welcome.

Enhanced Light
Crosswalk Illumination

UW–Milwaukee · UW–Madison · TAPCO

Pedestrians are hardest to see when they need it most

Deaths peak at night

Existing treatments fall short

Three gaps in guidance

SafeWalk™ prototype with seven adjustable spectra

281 nighttime rounds at two Wisconsin sites

Williamson St & S Dickinson St

N Maryland Ave · UWM campus

What we varied

Instrumentation

Measured illuminance at the crosswalk (Madison)

Field effort under harsh and tough conditions

Deep-learning eye tracking of real drivers

Collect

Synchronize

Detect

Measure

Reaction Time

Detection Distance

Quality control

See the system in action

Live demonstrations for the WisDOT POC

Blue-spectrum lighting won at both sites

Madison · Blue Baseline

Milwaukee · Ice Blue

Spectrum comparison

Recommended Milwaukee configuration

Spectrum comparison

Recommended Madison configuration

Cross-site synthesis

Eye-tracking physiological metrics (N = 163)

Final recommended lighting specifications

From two sites to any Wisconsin crosswalk

Blue Baseline OLS models

Stopping sight distance screen

Required checks before installation

Implementation recommendations

Questions, corrections, and suggestions