content

A single-object video tracking dataset

Content

>Tracking dataset

This section presents, via the left menu, a description of the test sequences for each modeled situations along with frame samples, low resolution video previews and the event annotations. Annotations have been done using the VIPER toolkit and modified later.

Modeled situations
Covered tracking problems
Ground-truth annotation format

Modeled situations

Currently, the dataset contains 126 sequences related with the single-object video tracking task (with around 22000 frames annotated). These sequences represent the common problems in video tracking in different testing situations. We distinguish four situations:

Situation 1: synthetically-generated sequences with isolated tracking-related problems and varying degree of complexity.
Situation 2: sequences recorded in a controlled environment (laboratory) with isolated tracking-related problems and different degrees of complexity.
Situation 3: selected sequences from public datasets. They are classified according to their tracking-related problem (only one problem is allowed per sequence), their target type (cars, people and faces) and their estimated complexity.
Situation 4: selected sequences from public datasets. They contain various tracking-related problems classified attending to their target type (cars, people and faces) and estimated complexity.

Sample frames from each situation are shown in the following images:

Situation 1:

Situation 2:

Level2

Situation 3:

Level3

Situation 4:

Level4

Covered tracking problems

Several tracking-related problems are included in the sequences in order to evaluate the adaptability of the algorithms to real life problems. They include complex movement of the target, global and local illumination changes, noise, occlusions, scale changes and similar objects in the background. For estimating the complexity of such problems in each sequence, some criteria have been defined (available here).

Complex Movement

ComplexMovement