I like biking. As a bicyclist, sometimes it’s not always safe to be driving around cars, especially on highways or roads without dedicated bike lanes.
What it does
IoT Bicycle Collision Detection helps solve the problem of bike safety by audibly alerting bikers when cars are approaching. The solution is a small Raspberry Pi-based analyzer that can take in video and send alerts via Bluetooth to tell a biker that cars are coming up nearby.
How I built it
I worked with the Raspberry Pi to analyze visible-light camera video. The video was run through
ffmpeg to break it into individual frames, which were then analyzed in turn. Each image was run through a trained machine learning model using gcp to detect features that resembled cars and vehicles on the roadway. Since frames were analyzed individually, the video could be streamed and processed in real time with low latency, allowing a biker to know quickly when there is danger. Frames were broadcast to a monitor over ssh for demo purposes at the same time audio was sent to the biker via Bluetooth.
Challenges I ran into
One big problem was I did not have a bike or good camera to test my setup on. I ended up using some other video to get started. Technically speaking, I had some trouble getting set up with some packages on the Raspberry Pi, as well as in Python.
Accomplishments that I'm proud of
I set up a cool system that reliably detects and alerts on car movements.
What I learned
I learned a bunch about working with vision.
What's next for IoT Bicycle Collision Detection
I want to try to implement this system in a live environment, where video would be captured from a Raspberry Pi mounted on a bike as it is moving.