Definition: Image tracking (image recognition and tracking) is an augmented reality method that finds a known, flat image in the camera image, calculates its position in space and anchors digital content to it. The image itself, for example a catalog page, a trade show poster or a label on the machine, becomes the anchor point for a 3D model, an animation or information.
Image tracking is a special form of marker tracking. Classic markers are patterns designed specially for machines, such as QR codes or ArUco codes. Image tracking, in contrast, works with natural motifs: photos, graphics and layouts that you print anyway. This entry focuses on these image targets.
In B2B mechanical engineering, image tracking is used for: product brochures with a 3D view, catalog pages as an entry point to variants, trade show posters with an animated machine, and nameplates and labels as an entry point to service content.
Advantage: Existing print media become an access point to interactive content, without a break in the design.
English: Image Tracking, Image Targets, Image Recognition
German: Bildverfolgung, bildbasiertes Tracking, Bildziel-Erkennung
How It Works
Features instead of patterns
Image tracking is based on computer vision. The software does not read a pattern as with a QR code. Instead, it compares distinctive image points:
- Prepare the reference image: The software analyzes the motif in advance. It looks for corners, edges and high-contrast details (feature points) and saves them as a compact description in an image database.
- Search the camera image: During use, the system looks for the same features in each frame.
- Calculate the position: When the software finds enough features, it calculates the position and orientation of the image in space (pose estimation).
- Anchor and follow the content: The digital content is aligned with the image and repositioned each time the camera or the image moves.
Role of the physical size
For a 3D model to appear true to scale, the system must know the real size of the printed image. In ARKit, you specify the physical size of each reference image. ARCore can also estimate and refine the size itself, but a specified size improves detection and tracking.
Interaction with spatial tracking
Modern frameworks combine image tracking with markerless tracking and SLAM. The image then serves as the starting point: after the content is placed, it stays stable, even when the image briefly leaves the field of view.
Image Tracking vs. Marker Tracking
| Criterion | Image tracking (image targets) | Classic markers (QR, ArUco) |
|---|---|---|
| Design | Natural motifs, fits into the layout | Abstract black-and-white pattern |
| Detection | Comparison of image features | Decoding of a pattern |
| Requirements | Much detail, no repetitions | High contrast, free margins |
| Number of variants | Limited by the image database | Very many unique IDs |
| Typical use | Marketing, print, packaging | Position reference in production and logistics |
A logo or a line drawing is a poor image target because it has too few features. In the ARCore documentation, Google explicitly advises against barcodes, QR codes, logos and line art as reference images.
B2B Applications
Sales: brochures and catalog pages
The field sales representative puts a product brochure on the table at the customer site. The tablet detects the page and shows the machine as a 3D model above it. Each catalog page can open its own model or variant. The printed material stays the guide for the conversation, and the 3D view supplements it.
Trade show: posters and booth graphics
Large-format posters at the booth become entry points: visitors point a tablet or smartphone at the motif and see the system in operation, for example an animated material flow. For large image targets, specifying the real size improves stability. You can reuse the content after the trade show in mailings and ads.
Service: nameplates and machine labels
A label or sticker on an assembly serves as the anchor point for maintenance instructions. The technician scans the label and sees exploded views, spare parts information or step-by-step instructions on the correct component. Important: engraved or shiny metallic nameplates are often poor targets. A printed, matte label with enough image detail works more reliably.
Training: manuals and learning cards
Training documents contain illustrations that show the applicable 3D model when scanned. Participants rotate assemblies, hide housings and understand functional principles before they work on the real machine.
Design review: drawings as anchors
A printout of the hall layout or of an assembly drawing on the meeting table serves as a common reference point. All participants see the model on their devices at the same location and at the same scale.
Requirements for Good Image Targets
- Rich in detail: Many irregular edges and corners, for example photos with texture.
- High contrast: Colors are almost unimportant. Brightness contrast is what matters. ARCore evaluates reference images without color.
- No repetitions: Grids, tile patterns and large single-color areas confuse the feature search.
- Flat and matte: Curved, creased or reflective surfaces interfere with detection.
- Large enough in the image: For detection, ARCore requires the image to fill at least 25 percent of the camera image.
- Check before printing: Tools such as arcoreimg (score from 0 to 100, at least 75 recommended) or the image target rating in commercial SDKs show the suitability before printing.
Frequently Asked Questions (FAQ)
How is image tracking different from marker tracking?
Image tracking is a subtype of marker tracking that uses natural images instead of abstract codes. Classic markers are more robust and distinguish very many IDs, but they look technical. Image targets fit into brochures and posters, but they set higher requirements for the motif.
How many images can an application detect?
That depends on the framework. ARCore allows up to 1,000 reference images per database and tracks up to 20 images at the same time. ARKit also limits the number of images tracked at the same time. For large catalogs, image databases are often split by chapter or campaign.
Does image tracking need an internet connection?
Not necessarily. ARCore detects and tracks images completely on the device. This is relevant for trade show halls with weak Wi-Fi and for service work in production areas. Cloud-based image recognition is an option for very large image databases.
Does image tracking work in the browser?
Partially. Until now, the WebXR Device API offers image tracking only as an experimental feature that is not available in all browsers. Web AR libraries therefore use their own image recognition with JavaScript and WebAssembly. For production scenarios that require high stability, native apps based on ARKit and ARCore are usually more reliable.
Does image tracking also work with headsets?
Yes, depending on the platform. On visionOS, ARKit provides a dedicated interface for image tracking. With a headset, your hands stay free, which is an advantage for service work on the machine.
Industry Standards and Technologies
- ARCore Augmented Images – image tracking on Android, including requirements for reference images
- ARKit: Detecting Images in an AR Experience – image detection on iOS and iPadOS
- ARKit ImageTrackingProvider – image tracking on visionOS
- Vuforia Image Targets – image targets in a widely used enterprise AR SDK
- WebXR Marker Tracking (Incubation) – draft for image and marker tracking in the browser
- ISO/IEC 18039 – reference model for mixed and augmented reality
Related Topics
Tracking methods: - Marker tracking – tracking with reference markers - Markerless tracking – tracking without markers - SLAM – mapping and localization - Object tracking – detection of real 3D objects - Plane detection – detection of planes - Feature detection – search for image features
Platforms: - ARKit – Apple's AR framework - ARCore – Google's AR platform - visionOS – Apple's operating system for headsets - HoloLens – mixed reality headset - WebXR – AR and VR in the browser
Basics: - Augmented reality – digital content in the real view - Mixed reality – merging of real and digital - Extended reality (XR) – umbrella term - Sensors – camera and motion capture - Computer vision – machine vision
3D content: - 3D model – digital geometry - 3D pipeline – from CAD to real-time display - Real-time rendering – real-time graphics - USDZ – 3D format for Apple platforms - WebGL – 3D in the browser
Applications: - Print AR – print media with a 3D extension - AR maintenance instructions – service on the machine - Immersive experience – immersion in content - 3D product configurators – interactive variants
