Overview

The Smart Zoom node allows for panning and zooming to cover specific objects detected by any AI model. This functionality is crucial for applications requiring focused attention on particular objects within a video frame.

Inputs & Outputs

Input media: Raw Video.
Output media: Raw Video.
Output metadata: None.

Pipeline usage

Place a detector, and normally Track Objects, before Smart Zoom. The node builds one view that contains all matching object boxes, then gradually adjusts toward it; multiple distant objects therefore produce a wider view than one isolated object. zoom_adjust_threshold suppresses small changes, and zoom_reset_interval returns to the full frame after the configured number of frames without a match. Because this node changes pixels but does not transform existing object coordinates, use it near the end of a branch, after analytics and before display or encoding.

Properties

Property	Description	Type	Default	Required
`object_types`	ex. `car`,`person`,`face.mask`. Zoom to focus on detected objects with specified types. Use the `object_label.class_label` format for objects that also have a specific class.	model-labels	`null`	No
`zoom_adjust_speed`	Larger is faster. Range: minimum `0`, maximum `100`. Step: `5`.	`integer`	`10`	No
`zoom_adjust_threshold`	Don't adjust zoom unless a new zoom window differs from current one by more than this percentage. Range: minimum `0`, maximum `100`. Step: `5`.	`integer`	`15`	No
`zoom_reset_interval`	Reset zoom if no objects matching type are seen for this time. Range: minimum `0`, maximum `100`. Step: `5`. Unit: frames.	`integer`	`15`	No

Output Metadata

Node does not insert any output metadata; upstream metadata continues through the pipeline