Computer vision

Use these guides to add vision capabilities to your machine. The standard pipeline is: configure a camera, deploy an ML model, add a vision service, then use detections or classifications in your code.

Not sure whether to use detection or classification? Detection finds where objects are in an image (bounding boxes). Classification tells you what the whole image contains (labels). Use detection when you need object locations; use classification when you just need to categorize the scene.

Configure a vision pipeline

Deploy an ML model and configure a vision service to give your machine the ability to detect or classify objects.

Detect objects

Retrieve 2D bounding-box detections from a vision service, filter by confidence and class, and run detections in a loop.

Classify images

Use a vision service to classify images by label and confidence, make decisions based on classification, and monitor scenes continuously.

Track objects

Match detections across consecutive frames using IoU, assign stable track IDs, and count entries and exits.

Measure depth

Retrieve point clouds and depth images from a depth camera, read depth at specific pixels, and measure distance to detected objects.

Act on detections

Build a module that uses vision service results to control machine behavior.

Alert on detections

Send email or webhook alerts when your vision service detects specific objects or classifications.