Overview
This project combines computer vision, vehicle telemetry, and embedded Linux into a single in-vehicle AI system. An NVIDIA Jetson Nano Super runs a live YOLO object detection model on a dashcam feed while simultaneously pulling real-time vehicle data over USB OBD-II — all written in Python and running on a fuse-tapped power circuit inside the car.
The core idea was to fuse two independent data streams — what the camera sees and what the engine reports — to create a smarter dashcam that reacts to context rather than just recording all the time.
What I built
The system runs on a Jetson Nano Super mounted in the vehicle. Key subsystems:
- Event-triggered recording: Custom trigger logic monitors the YOLO detection stream and OBD-II readings simultaneously. Recording auto-initiates on collision events, hard braking, or aggressive driving patterns detected through fused sensor data.
- Predictive maintenance: Anomaly detection on live OBD readings flags unusual engine behavior — RPM spikes, coolant temp excursions, fault codes — and logs them with timestamps for later review.
- Voice command interface: A lightweight voice interface lets me issue commands hands-free while driving — start/stop recording, query vehicle status, or trigger a manual save.
- Wireless video transfer: Recorded clips transfer wirelessly to a mobile device over the local network without pulling the unit from the dash.
- In-vehicle power: The Jetson is powered via a fuse-tapped cigarette lighter circuit I wired myself — clean installation with no dangling adapters, handling all hardware integration and electrical work personally.
Outcome
The system is deployed and running in my vehicle. It's been the most technically demanding project I've taken on — it required pulling together embedded Linux, computer vision, real-time serial comms, audio processing, and automotive electrical work into one coherent system.
The project directly reflects skills I want to bring to engineering work: systems thinking, Python, embedded hardware, and the ability to take something from idea to deployed hardware.