Object Tracker

Overview

The Object Tracker plugin is a model plugin that propagates bounding boxes forward through video frames.

Use this plugin when you have manually marked an object in one frame and want the plugin to create predicted bounding boxes for the following frames.

Object Tracker

Plugin Functionality

The Object Tracker plugin runs on video assets. It looks for bounding-box annotations whose class title ends in _toTrack, tracks those boxes forward through the video, and returns predicted bounding boxes using the matching class without _toTrack.

For example, if your ontology contains:

• Vehicle

• Vehicle_toTrack

you can draw a Vehicle_toTrack bounding box on a video frame. The plugin will create predicted Vehicle boxes on later frames.

The plugin only processes bounding-box annotations with class titles ending in _toTrack. Make sure the project ontology also contains the matching destination bounding-box class without _toTrack.

Supported Data Types

• Video

Supported Annotation Tools

• Bounding Box

Plugin Configuration

You may vary a number of settings related to tracking from the Config JSON field. Each option is detailed below:

{
  "model_index": 4,
  "num_predictions": 20,
  "stride": 1,
  "include_key_frames_only": true
}

• "model_index": Selects the OpenCV tracker used by the plugin.

  • Options:

    • 0: BOOSTING

    • 1: MIL

    • 2: KCF

    • 3: TLD

    • 4: MEDIANFLOW

    • 5: MOSSE

    • 6: CSRT

  • Example:

    • "model_index": 4

• "num_predictions": Maximum number of future frames for which the plugin should attempt to create predictions.

  • Example:

    • "num_predictions": 20

• "stride": Controls how often predicted boxes are returned. A stride of 1 returns every predicted frame, while a higher value skips intermediate predictions.

  • Example:

    • "stride": 1

• "include_key_frames_only": Included in the default configuration for consistency with video plugins. Object Tracker follows the frames it processes from the source video.

  • Example:

    • "include_key_frames_only": true