Vehicle detection

CCEM

Shows how many cars are in your city, where they drive, where they park, and whether it is getting more or less.

Aerial image of Hamburg with detected cars marked in yellow and delivery vehicles in green

The challenge

Traffic and parking are still counted by hand or with fixed cameras. Both are expensive and do not scale. Counting from the air failed on the conditions of the city: tall, dense buildings that block the view to the ground, deep shadows in which a black car disappears, and so many objects so close together that a car is hard to tell from a container.

Our approach

CCEM solves this for the first time. It detects vehicles from above even under urban conditions, so cities can see how traffic is distributed, derive vehicle numbers, tell cars from delivery vehicles, and even map the parking space in the city.

Four questions every city should be able to answer.

Hamburg, 2018 to 2022, from the city’s own aerial imagery.

Map of Hamburg with a grid coloured from white to dark red by vehicle concentration — 2
Where are they?
The counts are aggregated per district or grid cell, so planners see how traffic is distributed across the city.
Vehicle concentration, Hamburg, 2022, from white (low) to dark red (high)

Map of Hamburg with a grid coloured from green to red by change in vehicle numbers — 3
Is it getting more or less?
Archive images reveal the trend, and show what a car-free zone or new parking fees actually changed.
Change in vehicle numbers, Hamburg, 2018 to 2022, from green (decrease) to red (increase)

Parked vehicles detected along a residential street in Hamburg — 4
Where do they park?
Parked cars reveal the parking space. CCEM maps every bay, marked or not, down to the individual space, and combines several years to even out construction sites and seasonal effects: a complete inventory of the city’s parking.
Parked vehicles detected along a residential street, Hamburg

Built for the city.

Standard vehicle detection is trained on open roads and parking lots. We trained CCEM for the conditions of the city.

150,000 vehicles, annotated by hand

Instead of relying only on standard datasets of open roads, we annotated 150,000 vehicles in 40 square kilometres of Hamburg’s city centre, in shade, between trees and under canopies.

Up to 92 % detection rate

On aerial images CCEM detects up to 92 % of all vehicles, and up to 15 % more than a model trained on a standard dataset alone.

Works on archive images

Most cities fly aerial surveys every few years. CCEM reads those archives, so the trend is available from day one, without new flights or new sensors.

Where CCEM helps

Objective vehicle counts across the whole city
Traffic distribution by district
Evaluating car-free zones, parking fees and other measures
Parking space detection and parking pressure
A fact base for planning and public communication

Who CCEM is for

Cities and municipalities
Urban and transport planners
Parking and mobility operators
Transport authorities and research institutes

Developed in 2023 with funding from the mFUND programme of the Federal Ministry for Digital and Transport. Imagery: Freie und Hansestadt Hamburg, Landesbetrieb Geoinformation und Vermessung.