← Major ProjectsIn Progress — Summer 2026
I got hit from behind by a car while cycling. Cyclists almost never see what's coming — by the time you hear it, it's already close. I wanted to build something that buys a few extra seconds.
The idea: a rear-facing camera on the bike that detects approaching cars in real time and triggers a haptic buzz on the handlebars before the car gets dangerously close. No phone, no display — just a physical alert you feel through your hands.
What It Does
Detects
Rear-facing camera feeds into a semantic segmentation model (FasterSeg) running on-device at ~15 FPS. It identifies cars, trucks, and motorcycles in the frame.
Tracks
A Kalman filter tracks each detected vehicle across frames and estimates how fast it's closing in — not just "car present" but "car approaching fast."
Alerts
When something crosses the threat threshold, a GPIO pin fires a MOSFET that drives a haptic motor mounted on the handlebar. The buzz pattern scales with urgency.
Software Stack
Model
FasterSeg — lightweight real-time segmentation
Runtime
ONNX Runtime C++, INT8 quantized
Tracking
OpenCV Kalman filter
Alert
RPi.GPIO + MOSFET driver
Pipeline
3 threads: capture / infer / alert