Week 9 Motion Analysis and Video Segmentation Flashcards
(14 cards)
What does motion analysis study?
Motion analysis studies the movement of objects within a video or image sequence.
What is optical flow?
Optical flow estimates the apparent motion of pixels between consecutive frames.
What is the difference between dense and sparse optical flow?
Dense optical flow computes motion vectors for every pixel. Sparse flow tracks key points.
What is Lucas-Kanade used for?
Lucas-Kanade is a popular algorithm for sparse optical flow estimation.
What does motion segmentation achieve?
Motion segmentation separates moving objects from the background in a scene.
What is background subtraction?
Background subtraction is a common technique for motion segmentation.
What are applications of motion analysis?
Applications include video surveillance, autonomous driving, and activity recognition.
What is video object segmentation?
Video object segmentation identifies and tracks objects in a video over time.
How do semi-supervised segmentation methods work?
Semi-supervised segmentation uses an initial object mask for tracking.
What are challenges in video segmentation?
Challenges include occlusions, varying object appearances, and motion blur.
How is optical flow represented?
Either using a 2D vector field or a colour coded map to show magnitude and direction of motion.
What is the circulant matrix method?
It’s an alternative approach to a motion tracking classifier.
It stores features of the training image and all potential shifts.
It is then used as a kernel over potential regions that may contain object of interest.
How does Learning Correspondence from the Cycle-Consistency of Time work?
Given unlabeled data it takes an initial patch.
Tracks it backwards and then tracks it forwards.
The inconsistency is then used as the learning signal.
It then learns to minimises the tracking inconsistency.
