Count every bike that passes.

Open-source software that turns street-camera footage into reviewed counts of cyclists.

Download the appRead the source

Windows, macOS and Linux

The review screen of the app: a list of passages, the clip of a cyclist crossing the counting line, the suggested class and a timeline of every crossing.
The review screen. Every passage gets a clip, a suggested class and a place on the timeline.

Five stops from footage to count

The software proposes. A person decides.

Any fixed camera will do

Open a file from your computer or paste a Drive link. Group the hours of one street together.

The library of the app with groups of street videos.

A line where people cross

Draw counting lines at any angle and choose what each one counts: the bike lane, the sidewalk, the road.

The line editor with counting lines drawn over a frame of the street.

Everything that moves is a candidate

Object detection and motion tracking find each passage. A vision model suggests its class, blurred night riders included.

Frames of one passage, cut around a cyclist, as the vision model receives them.

Accept, correct or add by hand

Enter accepts a suggestion and a number key changes it. Missed someone? Shift and a number counts them while the video plays.

The original video in the review screen, with the counting lines on top and the bar for counting by hand.

A video and a spreadsheet

Get the footage with a running counter burned in, a CSV of every passage and a JSON report.

A frame of the final video with the counter panel over the street.

Checked against counts made by hand

Detection is the first pass. The review closes the gap.

Two streets, one evening each, compared with counts made by hand
WhatStreet and hourBy handBefore review
CyclistsAndradas, 5 to 6 pm7063
PedestriansAndradas, 5 to 6 pm8560
CarsAndradas, 5 to 6 pm171185
CyclistsAugusto de Lima bike lane, 5 to 8 pm3529

What the detector missed is added by hand. What it doubled is discarded.

Cities build for what they count.

counter.bike is working toward one open count of cyclists, worldwide. It starts with your street.

Four ways to help

The code, the models and the method are public.

It runs on your own computer

Systems
Windows, macOS and Linux
Graphics card
NVIDIA is faster. A processor alone works.
Vision model
Your own OpenRouter key, about 18 US cents per thousand passages
Licence
AGPL-3.0, free to use
Download for free
A frame of the final video: the street with the counter panel showing totals per category.
A frame of the final video, with the counter burned in.