SLEEP
PIR ARMED
DEVICE CAMERA
AI roadside wildlife detection · live simulation

It sees the deer before the driver does.

A solar-powered roadside unit that spots wildlife, flashes a warning to oncoming drivers, and steers the animal toward a safe crossing with a species-tuned tone.

Take control Propose a pilot corridor
With WildSight

A deer is heading for the road.

  1. Motion
  2. AI ID
  3. Driver beacon
  4. Tone
  5. Safe crossing
0conflicts with WildSight0 runs
0conflicts today0 runs
2nd place · Devpost GPT-4o Code & Conquer AI Hackathon 2 working prototypes · AI vision + frequency emitter University of Florida · Engineering Innovation Team 7 Scripted simulation · illustrative
Deployment planner

Where the first units would prevent the most crashes.

The road network of eight North Central Florida counties, ranked by expected animal–vehicle crashes from 2014–2024 crash reports, traffic and habitat. Choose how many miles to cover and see the units, LoRa gateways, cost and crashes avoided.

The page scrolls over the map; Ctrl + scroll or pinch zooms it, and two fingers pan it on a touch screen. Open the planner full screen ↗ Planning estimates
The problem

Wildlife collisions cost the U.S. $8.4 billion a year, and the signs aren't working.

Most collisions happen on ordinary two-lane roads, far from the few places that have fences or overpasses. The standard fix, a static yellow sign, is where almost none of the crashes happen.

1–2M

collisions between vehicles and large animals every year in the U.S.

FHWA Report to Congress
$8.39B

estimated total annual cost: repairs, medical care, towing, carcass removal, animal value.

FHWA
~200

people killed and about 26,000 injured each year.

FHWA
2%

of animal–vehicle crashes happened within sight of an animal-crossing sign (Utah study).

Khalilikhah & Heaslip, 2017

65.2% of the 45 people we surveyed have witnessed an animal collision. Armadillos, raccoons and squirrels topped the list, followed by opossums, deer and birds. Over half rated their concern for wildlife populations 4 or 5 out of 5. Team 7 Animal Collision Survey

How it works

Sleep, see, decide, warn, record.

WildSight spends almost all of its life asleep. It only spends energy, and only makes noise, when an animal is actually there. Click a step to replay it in the simulation at the top of the page.

The unit

Built from parts that already work.

Motion sensing, computer vision and frequency emitters are proven technologies. WildSight packages them in a weatherproof, solar-powered case that straps to a stake, a guardrail or a sign post, with service about once a year.

Hover a part to highlight it, explode the case, or switch the mount. The design follows the team's SolidWorks case, 3D-printed on a Bambu Lab printer.

drag to rotate
Simulation Lab

Now run it yourself.

A live 3D roadside with real traffic. Pick an environment, switch WildSight on or off, and turn individual features off to see what each one contributes. Drag to orbit; Ctrl + scroll to zoom. Animals arrive on their own every few seconds.

Environment
WildSight ON Beacon off Tone —
SLEEP
PIR ARMED
DEVICE CAMERA
Sleep
Motion
Classify
Beacon
Tone

Outcomes Simulated

0detections
0redirected to crossing
0crossed, no conflict
0near misses + collisions
Battery 100% · solar charging off (night)

Unit controls

Event log

    Outcome odds in this simulation are illustrative and set so the effect of each feature is visible. WildSight's real effect on animal behaviour will be measured in the pilot. Detection thresholds (50% / 75%) and species frequencies come from the team's prototype and cited hearing-range research.

    Feature demos

    Each part of the unit, up close.

    Five small simulations, one per capability. The hardware itself is at the top of the page: rotate it, explode it, change the mount.

    AI species ID

    What the device camera sees at night, and how confident the model is. Try an object it wasn't trained on.

    Simulated output

    Species-tuned frequency

    Every species hears a different band. WildSight picks a tone near the top of the detected animal's range, above human hearing.

    Cited data

    Hearing ranges: Columbia Hearing; LSU "How well do dogs and other animals hear?" Preview shifts the tone into human hearing with the prototype app's mapping (200 + (kHz − 0.3) / 84.7 × 1800 Hz).

    Solar power budget

    Why motion wake matters: a camera that never sleeps outruns a small panel within days.

    Design assumptions
    —Wh in / day
    —Wh used / day
    —lowest charge, 7 days

    Assumed: 6 W panel at 75% efficiency, 40 Wh battery, 0.06 W sleep draw, 2 W camera + inference, 1.5 W tone, 1.5 W beacon; 6 s camera / 3 s tone / 10 s beacon per detection. Replace with measured figures from the hardware build.

    Corridor mesh · roadmap

    Units along a road tell their neighbours when one detects an animal, so beacons light up before drivers reach the spot.

    Planned feature

    From the team's SIT "multiplication" concept: units embedded along lamps, signs and guardrails form a network. Spacing shown is illustrative.

    Driver beacon: the warning arrives earlier

    A driver who sees an amber flash starts braking before the animal is in the headlights. Compare stopping with and without the warning.

    Illustrative physics

    Assumes 1.5 s perception–reaction, 0.7 g braking on dry pavement, animal first visible 60 m ahead in low beams, beacon visible 150 m ahead. Stopping distance = v·t + v²/(2·a).

    Research

    45 surveys, 30 interviews, two prototypes.

    Primary research by Engineering Innovation Team 7 shaped which animals the model covers, how the unit is powered, and what we removed.

    Who we heard from

    Witnessed an animal collision65.2%
    Identify as drivers97.8%
    Identify as pet owners84.4%
    Pet owners worried about their pets (3–5 of 5)70%+
    Concerned for wildlife populations (4–5 of 5)50%+
    Survey, n = 45. Most respondents were 18–24.

    30 stakeholder interviews

    Animals most often seen in collisions:

    TIER 1ArmadilloRaccoonSquirrel
    TIER 2OpossumDeerBirds

    Drivers told us they'd pay $100–300 for a wildlife detection system. Insurance agents and repair techs pointed to rural roads at dawn and dusk, in spring and fall.

    What the AI was trained on

    4,665 Kaggle images, transfer learning on MobileNet via Google Teachable Machine, exported to TensorFlow.js. Raccoons are a top-3 collision species with the thinnest data: the first thing to fix.

    What testing changed

    Removed after peer review: pheromone repellent (refills), AC power (cost), strobe light (driver distraction, now a softer blinking beacon), mounting on crossing signs (could disrupt migration).

    Next model: add people, vehicles and a "no animal" class; classify live video rather than stills.

    Next emitter: directional speaker aimed at the field, so nearby homes and drivers aren't exposed.

    Prototype web app identifying an armadillo with its audio range
    Prototype 1 · AI detection appUpload or camera capture, species name and hearing range, tone when confidence is over 75%, timestamped log.✓ Identified trained species with high accuracy! Always picked one of its 8 classes, even for a personTry the demo · Watch video
    3D-printed blue WildSight case prototype
    Prototype 2 · frequency emitter + caseArduino generating four button-selected frequencies, checked with a pitch detector. Case modelled in SolidWorks and printed on a Bambu Lab printer.✓ Output within 5–10 Hz of target (pass criterion ±20 Hz)! Test tones were audible: next version needs a directional emitterWatch device demo
    Competitive landscape

    Nothing else detects, identifies and warns every driver from the roadside.

    Car-based systems protect one vehicle. Crossings protect one spot. Whistles have little evidence behind them. WildSight sits in the gap: cheap enough to spread along a corridor, smart enough to act per species.

    SolutionDetects animalsIdentifies speciesWarns all driversActs on the animalOff-grid roadsideCost to deploy
    WildSightpilot will testLow, per unit
    Animal-crossing signstaticVery low
    Wildlife overpass (ARC Solutions)physical path—$5–10M each
    Save-A-Deer whistleweak evidence—Very low, per car
    Yard Sentinel repellerPIRmanual dialLow, not road-rated
    Garmin AI dash camone car—High, per car
    Tesla driver assistancenot wildlife-tunedone car—Built into vehicle
    Yes Partly / unproven NoAssessment from the team's competitive analysis and white-space matrix.
    Market & go-to-market

    Start on 880 miles in Alachua County. Scale to 3.9 million.

    Beachhead: Alachua County, Florida, with the Florida DOT. Then state DOTs, private road operators, conservation NGOs and local governments near crossing hotspots.

    $350M

    Federal Wildlife Crossings Pilot Program, FY2022–26; at least 60% for rural projects. Eligible uses include technology demonstrations.

    USDOT
    82%

    average drop in large-animal collisions at seven Swiss sites with animal detection + driver warning systems.

    FHWA 2008
    $2.5B

    projected wildlife conservation technology market by 2028, 8.2% annual growth.

    Industry reports compiled by team
    $40K

    cost of a single collision with a large animal like elk or moose, up to.

    MDPI sensors review
    Who it's for

    One unit, five groups who come out ahead.

    Roadmap

    From prototype to a standard roadside device.

    Stage 1 · now

    Develop & test

    • Functional unit with core features
    • Controlled tests: detection, deterrence, driver alerts
    • Refine per species with conservation experts
    • Open talks for regulatory approval
    Stage 2

    Pilot & awareness

    • Install units on a rural corridor in Alachua County
    • Measure animal behaviour and collisions
    • Public awareness with local authorities
    • Share pilot data with agencies
    Stage 3

    Commercialize

    • Production for state DOTs and private road operators
    • Work toward a standard collision-prevention device
    • Open sales to private landowners
    Stage 4

    National rollout

    • Installations in high-risk regions nationwide
    • Fleet data improves the AI models
    • Long-term ecological impact studies
    • Regular software and hardware updates
    Team

    Six engineers who built it before pitching it.

    ACAllario CarniIdeation & design
    ACArianna CortésProject management
    JJJennifer JuarezManufacturing
    SLSean LucenteMaterials
    JRJay RosenAI development
    RSRafael SevillaSoftware development
    The ask

    Pilot partners and seed funding to take WildSight from prototype to a field-tested unit.

    We're looking for a DOT or county partner for a pilot corridor, a conservation partner to co-design the behaviour study, and funding for a ruggedized hardware run.

    Roles we want to add:

    Animal specialistsElectrical engineersData scientistsGovernment relationsField installationLegal & IPSupply chainMarketing
    Source code & docs GitHub Archived release DOI 10.5281/zenodo.14739247 Live AI prototype Web app
    Start a conversation

    Copy this note and send it to the WildSight team via the GitHub repository above.