Overview

Use the Track Objects node after a AI Model Node to track any detected objects across the video frames. This node assigns unique identifiers to each object and persists them across video frames.

These unique identifiers can be used, for example, to :

determine the presence of a new object in the video
count the number of unique objects crossing an imaginary line in the scene
determine the velocity of an object (or aggregate objects)
determine how long a particular object has been in the scene (for instance, to detect loitering)

Caveats: The tracker only tracks objects that continue to appear in consecutive video frames - so if an object disappears from the video and then reappears some time later, it will be tracked as a new object at that time.

This node adds tracking IDs to the metadata; these may be displayed on the stream using the Overlay Node or processed using the Function Node.

Inputs & Outputs

Inputs : 1, Media Format : Raw or Multiplexed Video
Outputs : Raw or Multiplexed Video (same as Input format)
Output Metadata : Object ID

Properties

Property Value

Tracker Accuracy high ( uses dcf tracking method) : This is the most accurate, but computationally intensive tracking method. Highly robust against partial occlusions, shadow, and other transient visual changes. Less frequent ID switches.

medium (uses kit tracking method) : Works reasonably well for simple scenes. Susceptible to change in the visual appearance due to noise and perturbations, such as shadow, non-rigid deformation, out-of-plane rotation, and partial occlusion. Cannot work on objects with low textures.

low (uses iou tracking method) : Least computationally intensive tracking method. No visual features for matching, so prone to frequent tracker ID switches and failures. Not suitable for fast moving scene.

Property	Value
Tracker Accuracy	`high` ( uses `dcf` tracking method) : This is the most accurate, but computationally intensive tracking method. Highly robust against partial occlusions, shadow, and other transient visual changes. Less frequent ID switches. `medium` (uses `kit` tracking method) : Works reasonably well for simple scenes. Susceptible to change in the visual appearance due to noise and perturbations, such as shadow, non-rigid deformation, out-of-plane rotation, and partial occlusion. Cannot work on objects with low textures. `low` (uses `iou` tracking method) : Least computationally intensive tracking method. No visual features for matching, so prone to frequent tracker ID switches and failures. Not suitable for fast moving scene.

Metadata

Metadata Property	Description
`objects[].id`	The unique identifier assigned to each object is added to the `objects` property added by the Model Inference Node