Please note: The algorithm descriptions in English have been automatically translated. Errors may have been introduced in this process. For the original descriptions, go to the Dutch version of the Algorithm Register.
Detecting people and objects in live drone footage
- Publication category
- Impactful algorithms
- Impact assessment
- DPIA
- Status
- In development
General information
Theme
Begin date
Contact information
Responsible use
Goal and impact
The algorithm instantly recognises people and objects in the images. It does this based on shape. When the algorithm was developed, each shape was linked to an object. A human would take longer to do this, and it would require more effort.
The police capture these images using a drone during emergency response operations. For example, to provide assistance or to apprehend suspects.
The impact on the general public is minimal. The algorithm does not use facial recognition or other physical characteristics. The police do not capture or view any more footage using the algorithm than they would without it. The algorithm helps the police to focus more effectively on the footage that really matters.
For the people captured on camera, however, the impact is significant. Police assistance is provided more quickly. And suspects can be apprehended more easily.
Considerations
With the help of the algorithm, the police can make better use of the footage. The algorithm only shows what it sees. It distinguishes between people and recognisable objects. To the algorithm, a human being is also an ‘object’ in the shape of a person.
It is important that we explain this clearly. The algorithm does not recognise people by name or face. It only recognises shapes. The camera operator (a police officer) monitors the feed live. They are able to recognise a person or object. As a result, someone can be identified.
The algorithm assigns a unique number to every recognised object. This makes it easier for the camera operator to distinguish between different people, vehicles or objects. In practice, the operator mainly tracks and recognises people and vehicles.
The greatest risk lies in communication: do we explain clearly what the algorithm does and does not do? The algorithm does not make decisions itself. Without action from the camera operator, nothing happens in response to what the algorithm sees.
Human intervention
The emergency response drone is operated by two police officers. The pilot controls the drone. The camera operator controls the camera and monitors the live footage.
The algorithm highlights whatever it recognises in the footage with a box. Each box is given its own number. The camera operator always checks for themselves what can be seen within that box. This makes it easier for the operator to assess the footage more effectively.
The algorithm’s recognition is only visible live in the footage. The algorithm does not control the drone or the camera. Unless the camera operator takes action, nothing happens in response to what the algorithm sees.
Risk management
Legal and ethical risks have been assessed in a GEB. A GEB is a privacy assessment. This assessment was specifically designed for the emergency response drone. We have reviewed the risks identified and, where necessary, accepted them.
From a financial perspective, we are purchasing a licence for the pilot. The costs are limited and the licence has a clear expiry date.
From a technical perspective, the algorithm has been assessed in terms of its architecture.
The algorithm runs solely on board the drone. This is known as ‘on edge’. No police data is sent externally to be processed by the algorithm. As a result, the risk of police data being lost is low.
Legal basis
The legal basis is Article 3 of the Police Act. The police use the algorithm on footage they capture whilst carrying out their duties. For example, when assisting victims, maintaining public order and apprehending suspects in the act. The footage is processed in accordance with Article 8 of the Police Data Act.
Links to legal bases
- Police Act 2012: https://wetten.overheid.nl/BWBR0031788/2026-01-23
- Police Data Act: https://wetten.overheid.nl/BWBR0022463/2025-07-01
Impact assessment
Operations
Data
This concerns indirect personal data. This refers to data relating to individuals, cars and other items. It is possible that this data could be traced back to an individual. However, this would require additional police work.
The algorithm has been trained to recognise shapes in the images as objects (people, cars, lorries). The MS COCO dataset was used for the training.
Technical design
The algorithm operates according to the YOLO principle. Images (in this case, live video) are scanned to recognise selected objects. These are then visually highlighted by a box bearing a unique serial number within the image.
The algorithm operates on-edge, meaning that no operational data is processed externally. Consequently, the algorithm is neither self-learning nor generative.
External provider
Similar algorithm descriptions
- An investigation by an expert examines whether certain digital images (photos and/or videos) were taken with a certain digital camera and/or mobile phone. To answer this question, the camera identification algorithm is deployed, which identifies the camera.Last change on 25th of June 2024, at 16:19 (CET) | Publication Standard 1.0
- Publication category
- High-Risk AI-system
- Impact assessment
- Field not filled in.
- Status
- In use
- This tool can be used to quickly search large amounts of video footage, such as that from surveillance cameras, for moving objects and certain movements. This helps to perform targeted searches through video footage, without having to look at everything manually.Last change on 15th of December 2025, at 16:28 (CET) | Publication Standard 1.0
- Publication category
- High-Risk AI-system
- Impact assessment
- Field not filled in.
- Status
- In use
- A tool that assists counter staff in verifying a person’s identity by comparing the face of the person presenting themselves with the photograph on an identity document. Helps to prevent look-alike fraud.Last change on 4th of June 2026, at 8:18 (CET) | Publication Standard 1.0
- Publication category
- Other algorithms
- Impact assessment
- DPIA
- Status
- In use
- This algorithm shields personal data in police data by automatically blinding vehicle windscreens. This is done when vehicle images are requested as part of an ongoing investigation.Last change on 31st of January 2025, at 13:04 (CET) | Publication Standard 1.0
- Publication category
- Other algorithms
- Impact assessment
- DPIA
- Status
- In use
- The local authority uses ANPR (automatic number plate recognition) cameras to check whether vehicles are permitted to enter a specific area (Environmental Zone). The camera reads the number plate; an algorithm then checks whether access is permitted. If this is not the case, a photograph is sent to the enforcement system, where a special investigating officer (BOA) assesses whether a fine or a warning should be issued.Last change on 8th of July 2026, at 11:58 (CET) | Publication Standard 1.0
- Publication category
- Impactful algorithms
- Impact assessment
- DPIA
- Status
- In use