Unsupervised Learning of Long-Term Motion Dynamics for Videos.
Zelun LuoBoya PengDe-An HuangAlexandre AlahiLi Fei-FeiPublished in: CoRR (2017)
Keyphrases
- unsupervised learning
- long term
- periodic motion
- moving camera
- short term
- dynamic scenes
- input video
- motion features
- recognition of human actions
- supervised learning
- video sequences
- static images
- image sequences
- key frames
- optical flow
- camera motion
- video frames
- space time
- spatial and temporal
- temporal domain
- semi supervised
- temporal coherence
- successive frames
- moving objects
- dynamical model
- human motion
- physical constraints
- motion estimation
- motion cues
- motion capture data
- video representation
- motion model
- traffic scenes
- video objects
- object recognition
- dynamic textures
- machine learning
- visual data
- motion analysis
- model selection
- reference frame
- expectation maximization
- neighboring frames
- recognizing human actions
- visual features
- dynamical systems
- video content
- video surveillance
- video clips
- motion parameters
- dimensionality reduction
- human activities
- adjacent frames
- multi view
- action recognition
- video data
- video stabilization
- high dimensional
- object motion