Object tracking follows instance labels through a series of frames in a video or image sequence.

By detecting and localizing objects in a frame, their position in subsequent frames is determined based on pixel information enclosed within the label boundary. All labels automatically created using interpolation are assigned a confidence score (α) of 99%.

Your video or image sequence’s pixel information is accessed by Encord.

Since tracking algorithms propagate labels forward through frames, object tracking works if labels are made on lower number frame (for example, near the start of the video).

Object tracking and interpolation are both techniques to automatically apply labels on a range of frames.

  • Interpolation applies labels in between reference labels.
  • Object tracking uses a single existing label and applies labels to frames following manual reference labels.

Ontologies

Ontology shapeInterpolationTracking
Bounding box
Rotatable bounding box
Polygon
Polyline
Primitive
Keypoint
Bitmask

* Polygon and Bitmasks use auto-segmentation tracking


Modalities

The following table shows the modalities that support auto-segmentation tracking.

Ontology shapeAuto-segmentation tracking
Images
Videos
Image Groups
Image Sequences
DICOM

Quick object tracking

  1. Create at least one bounding box instance label on the frame you want to start tracking the instance from.
More manual instance labels on different frames result in high object tracking accuracy.
  1. Right-click the shape .

  2. Click Track object. For bounding boxes and rotatable bounding boxes, select the Standard (fast) option to run auto-segmentation tracking for 30 consecutive frames.

Polyline, Keypoint, Object PrimitiveBounding Box, Rotatable Bounding Box

Keyboard shortcut

You can use keyboard shortcuts to run object tracking on polylines, keypoints, and object primitives.

  1. Click the instance label.
  2. Use the Shift + T keyboard shortcut to the run object tracking for 30 frames.
Using the keyboard shortcut on bounding boxes and rotatable bounding boxes runs auto-segmentation tracking.

Fine tune object tracking

  1. Create at least one instance label on the frame you want to start tracking the instance from.
More manual instance labels result in high object tracking accuracy.
  1. Click the Automated labelling button in the bottom left corner of the Label Editor to bring up options for automated labeling.

  2. The Tracking and interpolation section is open by default.

    • Under the Method heading, select Tracking.
    • Select the object instance you want to track.
    • The Tracking annotation interval lets you specify how many frames after a manual label the object is tracked for.
    • Tracking range specifies the total range of frames instances are tracked across.
    • The Advanced settings lets you choose between the Advanced and Standard tracking models. The Advanced model provides higher accuracy, but takes longer to run.
  3. Click Run tracking to track the instance.

Object tracking example

Lisa, an annotator, wants to track a blue car on all frames between frame 10 and frame 40 in a video.

  1. She draws a bounding box around an instance of a blue car on frames 10 and 40.
  2. She sets the Tracking range to [10,40] and the Tracking annotation interval to 30.
  3. She clicks Run tracking to label the object from frame 10 to frame 40.

Settings

The Object tracking section of the editor settings allows you to adjust the following.

Tracking range

The range that quick object tracking runs for can be adjusted in the Object tracking section of the editor settings.

The default tracking range is set to 30 frames.

Change tracking algorithm

You can change the tracking algorithm in the Object tracking section of the editor settings.

The choice of setting determines which algorithm is run when using the Shift + T keyboard shortcut.