Skip to main content

How to use Birdi’s Map Canvas AI Detect tool

Learn how to run AI detection on any map layer in Birdi, with zoom-based tile control for more accurate and flexible object detection.

Kayley Greenland avatar
Written by Kayley Greenland
Updated this week

Birdi’s Map Canvas AI Detect tool allows you to run AI detection on any visible map layer, not just processed orthomosaics. This includes:

  • Satellite imagery

  • Base maps

  • External WMTS layers

  • Uploaded raster layers

  • Large orthos where you want more control over detection behaviour

Unlike Full Ortho AI Detect, this tool gives you direct control over tile size, helping improve accuracy across a wider range of use cases.

When to use Map Canvas AI Detect

Use Map Canvas AI Detect when:

  • You want to detect objects on satellite or third-party imagery

  • You need more control over detection scale

  • Objects are being missed or incorrectly split using full ortho tiling

  • Features are very small or very large

  • You want to test detection without reprocessing imagery

This tool analyses what you see on the map canvas, rather than relying on automatically tiled orthophotos.

Step-by-step: Using Map Canvas AI Detect

1. Open your map

Load the map and turn on the layer you want to analyse (satellite, base map, ortho, or raster).

2. Activate AI Detect

Click AI Detect from the map toolbar and select AI tag map area.

3. Define the detection area

  • A dashed blue box shows the active detection region

  • Zoom and pan the map so the area of interest fits clearly inside this region

  • Make sure objects are clearly visible at your current zoom level

Tip: Detection works best when objects are visually distinct on screen.

4. Choose an AI model

Select the AI model you want to use (e.g. Birdi’s Basic Model (SAM3)).

If detecting stockpiles we recommend using Birdi Stockpile model, We have optimised this model to better detect the boundaries of stockpiles.

5. Set your zoom level (tile size control)

The zoom level acts as the tile size input for detection.

You’ll see guidance like:

Best detection range: 16–20

This is critical for accuracy.

Why zoom level matters

  • Too zoomed out → objects become too small and may be missed

  • Too zoomed in → large objects may be partially detected

Practical zoom level examples

Here are some real-world examples to guide tile size selection:

🌳 Trees (individual canopies)

  • Recommended zoom: 17–19

  • Helps separate overlapping canopies

  • Too zoomed out may merge trees together

🏠 Buildings / roofs

  • Recommended zoom: 16–18

  • Captures full roof geometry

  • Prevents roofs being split across tiles

🚗 Vehicles

  • Recommended zoom: 19–21

  • Higher zoom needed due to small object size

🛣 Road features / street assets

  • Recommended zoom: 17–18

  • Balances linear features without over-segmentation

6. Text-based detection

In the Text-based detection field:

  • Type the object you want to detect (e.g. tree, roof, vehicle)

  • The AI will search for matching objects within the detection area

7. Select output type

Choose how results are created:

  • Polygon – area-based features (trees, roofs, surfaces)

  • Marker – point-based detections

  • Volume – when volume calculations are needed

View guide on turning your AI detections into volumetric measurements ( You will need to ensure you have a DEM available in your map)

8. Adjust threshold confidence

Use the confidence slider to control detection strictness:

  • Lower values = more detections, lower confidence

  • Higher values = fewer, more accurate detections

9. Run detection

Click Detect objects.

Detected features will appear as a new layer in your map and can be:

  • Renamed

  • Styled

  • Measured

  • Exported

  • Used in reports

Best practice tips

  • Always zoom to the scale of the object, not the map

  • Start with a small area, then expand

  • If detections look wrong, adjust zoom level first, not confidence

  • Use Map Canvas AI Detect when full ortho detection:

    • Misses objects

    • Splits objects

    • Produces inconsistent results

Related AI Detect features

Did this answer your question?