Car detection

An algorithm that detects cars in Pléiades imagery.


The algorithm uses object detection in Pléiades display imagery to detect and quantify cars. The result is a GeoJSON file with points drawn on detected cars.

Car detection can be used for traffic and parking management, retail analysis, and urban planning.

An input Pléiades image of Berlin (Germany)

Detected car points drawn on the input image

See more on the marketplace.

Training data and accuracy

The algorithm has been developed using 180,000 images in 50 countries spanning 6 continents.

For optimal results, the algorithm must be applied to imagery captured over North America and Europe. You can apply it to similar areas, but the same level of performance isn’t guaranteed. The algorithm has limitations when dealing with highly shadowed imagery, imagery containing closely parked vehicles, and desert imagery.

Scenarios represented in the training data:

  • Different times of day
  • Different times of year
  • Different terrains
  • Different object configurations

Requirements for input imagery

Checkmark inline-icon The STAC item must be a display data product from Pléiades.

Checkmark inline-icon The STAC item must be CNAM-compatible. Check that the STAC item has been added to storage in 2023 or later.

Input parameters

Required parameters

Input imagery

You need to specify the STAC items you want to apply the process to.

Output title

You need to specify the title of the output objects. This title will be assigned to the resulting STAC item and STAC collection.

API input

Use the detection-cars-oi name ID for the processing API.

A sample input payload for the process


  "inputs": {
    "title": "Processing imagery over Berlin",
    "item": ""

inputs.titleobject / required
The title of the output objects: STAC item and STAC collection.
inputs.itemobject / required
The STAC item link in the following format:{collection-id}/items/{item-id}