DriveBy

Training the model
Use real world video captures as imput

Inspiration

The many knocked down, shared e-scooters & bikes in Berlin, strengthened by a recent German government decision to further allow their (better regulated) parking in pedestrian areas, but also looked into the wider market need for such a solution - like the added risk of follow-on accidents by fallen-over vehicles on walk- & cycleways, and the optical eyesore they create.

What it does

Detects knocked down scooters & bikes. The AI presumably identifies brands and positions/uprightness of vehicles, based on shapes, colors and saddle position.

How I built it

Azure Custom Vision trained models

Challenges I ran into

Find enough, freely available pictures of scooters/bikes lying around (nextbike was particularly hard).

Built With

Submitted to

Jelbi-Hackathon der BVG

Created by

I helped with some competitive analysis (i.e. that the in-built sensors in e-scooters new that the ideal vehicle would be shared bike fleets like Nextbike), qualitative & analytical feedback (i.e. why the AI may mistake an orange car’s t-shaped bumper grill for a Mobile), and image research (searching images of fallen TIER bikes).

I also helped position the need for safer e-scooter & bike sharing services, based on a recent decision by the German government to further allow their parking in pedestrian areas, but also looked into the wider market need for such a solution - like the added risk of follow-on accidents by fallen-over vehicles on walk- & cycleways, and the optical eyesore they create.

Thomas Siddons
Natively Quadrilingual Sales & Marketing Leader, Innovator and Advisor
Idea, realization.

Stefan S.