Study sets, textbooks, questions
Upgrade to remove ads
Lecture 8 - Visual Object Tracking, Introduction
Terms in this set (17)
List the two different types of tracking
1) Automatic control - Filter object state
2) Computer vision - Tracking by detection
Why is visual object tracking important?
Cue for behavior
Why is it important to create a generic tracking algorithm?
• Adaptation to environment
• Can handle interactions
What is the input to a VOT algorithm?
1) Image sequence (video)
2) Object bounding box in frame #1
What is the output of a VOT algorithm?
Bounding box for frames t > 1, determined from frames < t
What are some assumptions that are done in VOT?
1) Object is visible in all frames (at least partially)
2) Camera might be moving (no background model)
What are the two ways of doing VOT (detection)?
1) Generative approach (mostly used)
2) Discriminative approach
List some challenges of the VOT task
3) Changes in viewpoint
How do one measure accuracy in tracking?
1) IoU - Intersection over Union is often used
2) If IoU = 0, target is lost
When is a tracker robust?
When it has few losses of the objects
How is tracking speed measured?
Equivalent Filter Operations (EFO)
It's a hardware independent assessment of real-time capability
When the tracker drifts away from the target, we want the tracker to reinitialize to the target.
One way to do it is to reinitialize when the overlap is under a threshold.
One can also wait a couple of frames until reinitialization, due to occlusion.
What is commonly used metrics to measure how well a tracker performs?
Expected Average Overlap (EAO)
It accounts for both accuracy and robustness.
Average Overlap is computer per length and sequence.
Get the Expected Average Overlap curve by averaging over all sequences.
What's a classic performance trade-off in tracking?
EAO vs. EFO
Expected Average Overlap vs. Equivalent Filter Operations
What is the most common baseline tracker?
Sets with similar terms
Physics Chapter 2 (and a little Chapter 1)
Chapter 10 Object-Oriented Analysis and Modeling U…
Physics Chapter 3: Describing Motion
Sets found in the same folder
Lecture 2 - Feature Descriptors
Lecture 3 - Convolutional Neural Networks
Lecture 4 - Image classification with convolutiona…
Other sets by this creator
Other Quizlet sets
GRM 391 Final
Lab Quiz #2
Real Property Characteristics, Definitions, Owners…
Criminology Final Review 1st Sem.