Crash Scene 3D
Browser-based 3D crash reconstruction
A browser tool that builds a 3D illustration of a two-vehicle crash from a few inputs — the vehicles, their positions, speeds, headings, and the impact point — placed on the real roads where it happened, with a one-click PDF export. It is meant to illustrate what happened clearly. It is not an expert analysis or accident-reconstruction tool, and it does not produce evidence-grade conclusions.
Watch the Walkthrough
A start-to-finish walkthrough: entering the two vehicles, placing them on the real road, setting speeds and headings, marking the impact point, and exporting the PDF.
Try It Right Here
The live app, embedded below with a ready example loaded: play the replay, switch between the top, wide and driver cameras, or edit the scene. For a blank start and more room, open it in a new tab.
Inside an InsurifyAI Claims Call
crash-scene-3d was built for InsurifyAI, and the real use is hands-free: claims intake. A driver calls the InsurifyAI intake agent, built on ElevenLabs, and simply describes the accident. The agent takes the report by voice, checks the crash against the real road layout so it can actually be drawn, then texts the driver a private link to the exact 3D illustration shown above, with a one-page summary they can send to their adjuster.
- The call starts. The agent discloses it is an AI assistant and asks the driver to describe the accident.
- It takes the report. One question at a time: the place, both vehicles, which way each was going, the speeds, and where each one was hit.
- It checks the scene. It reads the report back, then checks the spot against the live OpenStreetMap layout so the crash can be drawn, fixing anything that does not line up.
- It texts the result. A private link to the 3D illustration and a one-page summary, with distances and a verification fingerprint, ready to forward to the adjuster.
claimlink--crash-scene-3d.netlify.app/s/4RrJhpUKC1sC4vFvFFOWD_qY
Illustration only, not a decision about fault.
From a Few Inputs to a 3D Scene
You describe the crash with a handful of values. The tool turns them into a 3D scene on the actual geometry of the road, then lets you export it.
Simple Inputs
Pick the two vehicles and set their positions, speeds, headings, and the impact point. No CAD, no modeling — just the facts you already know about the crash.
Real Roads
The scene is built on the actual road where it happened. An Overpass/OpenStreetMap query pulls the road network and a terrain pipeline grounds the geometry so the layout matches the real intersection or stretch of road.
3D Illustration
A WebGL scene rendered with react-three-fiber and drei shows the vehicles, their paths, and the impact in three dimensions — orbit, pan, and zoom to see the moment from any angle.
Map View
A react-leaflet map ties the 3D scene back to its real-world location so you can confirm you are reconstructing the correct spot before you build.
PDF Export
Export the illustrated scene to a PDF with pdf-lib — a shareable, printable summary of the layout for a file, an email, or a conversation.
Runs in the Browser
Everything happens client-side. There is no account and no upload step — open the page, build the scene, export the PDF.
How It's Built
A TypeScript + React single-page app. The 3D view is react-three-fiber/drei over Three.js; the map is react-leaflet. A pure-TypeScript physics and projection layer turns the inputs into positions and paths, a map pipeline fetches and grounds the road geometry, and pdf-lib handles the export — all in the browser.
Rendering
- react-three-fiber + drei (Three.js) for the 3D scene
- react-leaflet for the real-world map view
- Orbit / pan / zoom scene controls
Logic
- Pure-TypeScript physics for speed and heading
- Geographic projection from lat/lng to scene space
- Impact-point and path derivation from inputs
Data & Export
- Overpass / OpenStreetMap road-network queries
- Terrain pipeline to ground the geometry
- pdf-lib for the PDF export
Honest scope. Crash Scene 3D illustrates what happened from the values you give it. It is a visualization aid, not an expert witness — it does not perform forensic accident reconstruction and its output should not be treated as evidence-grade analysis.