CrashLens
Smart accident detection ecosystem
A Raspberry Pi 5 in-car device and app ecosystem that detects crashes, records evidence — video, sensor data and GPS — and dispatches automated alerts to drivers, first responders and insurance providers.
Problem
After a car crash, critical evidence — skid marks, sensor readings, exact impact force and location — disappears within minutes. No affordable in-car system existed to automatically capture, encrypt and deliver that evidence to the right parties.
Why it exists
Evidence at a crash scene has a lifetime of minutes. CrashLens exists to capture it automatically, at the moment of impact, and hand it to the people who need it — drivers, first responders and insurers.
What was built
A Raspberry Pi 5 unit with Camera Module 3, a 6DOF IMU, GPS and 4G LTE detects crashes, saves the video around the impact, encrypts the evidence and pushes it to the cloud. Native iOS and Android apps serve drivers and first responders; web dashboards serve admins and insurance teams.
System
- Camera Module 3Continuous capture into a ring buffer
- 6DOF IMUAccelerometer thresholds detect the impact
- GPSLocation attached to every evidence packet
- Raspberry Pi 5Python · detection, ring-buffer save, AES-256-GCM encryption
- Cloudflare WorkersHono.js API receiving uploads over 4G LTE / HTTPS
- SupabasePostgreSQL — incidents, users, roles
- Audit logImmutable chain-of-custody record
- Cloudflare R2Encrypted video evidence
- Driver appiOS (Swift) · Android (Kotlin)
- Responder appAutomated dispatch alerts
- Admin dashboardReact · role-based
- Insurance dashboardReact · role-based
- device — Pi 5 + IMU + GPS
- backend_api — Cloudflare Workers
- driver_app — iOS / Android
- responder_app
- admin_dashboard
- insurance_dashboard
- public_website
Technical decisions
Detect on the device
Crash detection runs on the Pi itself using accelerometer thresholds from the 6DOF IMU, so the moment of impact is recognised at the source.
Ring-buffer video
The camera records into a ring buffer. A detected crash triggers a save of that buffer, preserving the footage around the impact.
Encrypt before it leaves the car
Evidence is encrypted with AES-256-GCM on the device and pushed over 4G LTE via HTTPS.
Edge API, split storage
Cloudflare Workers (Hono.js) receive uploads. Structured data lands in Supabase (PostgreSQL); video lands in Cloudflare R2.
Chain of custody
An immutable audit log records the evidence trail so that it stays legally defensible.
One system, several surfaces
Native apps — Swift on iOS, Kotlin on Android — for drivers and first responders; role-based React dashboards for admins and insurers.
Outcome
- Real insurance-industry interest.
- The immutable chain-of-custody audit log keeps evidence legally defensible.
- Automated dispatch cuts first-responder notification time to seconds.