TITLE: Learning to Segment Moving Objects in Videos
AUTHOR: Fragkiadaki, Katerina and Arbelaez, Pablo and Felsen, Panna and Malik, Jitendra
FROM: CVPR2015
CONTRIBUTIONS
- Moving object proposals from multiple segmentations on optical flow boundaries
- A moving objectness detector for ranking per frame segments and tube proposals
- A method of extending per frame segments into spatial-temporal tubes
METHOD
- Extract motion boundaries by optical flow
- Generate segment proposals according to motion boundaries, called MOPs (Moving Object Proposal)
- Rank the MOPs using a CNN based regressor
- Combine per frame MOPs to space-time tubes based on pixelwise trajectory clusters
ADVANTAGES
- Using optical flow could reduce the noises caused by inner texture of one object. Optical flow is more suitable for detecting rigid objects.
- Using trajectory tracking could deal with objects that are temporary static.
- Segments are effective to tackle frequent occlusions/dis-occlustions.
DISADVANTAGES
- Too slow. Every stage would take seconds to process, which is not suitable for practical applications.
- Use several independent method to detect objects. Less computations are shared.
- The power of CNN has not been fully applied.
OTHER
- RCNN has excellent performance on object detection in static images
- For slidewindow methods, too many patches need to be evaluated.
- MRF methods neglect nearby pixels’ relation and could not separate adjacent instances.
- Methods of object detection in video could be categorized into two types i) top-down tracking and ii) bottom-up segmentation.