35
Topic regards: Browsing of Search Results Video Retrieval using Spatio-Temporal Object Tracking Face tracking Yuan-Hao Lai

Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

Embed Size (px)

Citation preview

Page 1: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

Topic regards:◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal◆ Object Tracking ◆ Face tracking

Yuan-Hao Lai

Page 2: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

Visual Islands: IntuitiveBrowsing of Visual Search ResultsEric Zavesky, Shih-Fu Chang, Cheng-Chih Yang Columbia University International conference on Content-based Image and Video Retrieval (2008)

Page 3: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Conventional and new approaches]

Page 4: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Visual Islands]

Page 5: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Significant Advantages]• Make local choices of features that is currently displayed, not all results.• Faster because far fewer sample• Original rank was determined by a direct query from the user instead of a ranking based on similarity

Page 6: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Axis Layout]

Page 7: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Visual Island / Non-Linear Navigation]

Page 8: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Speed & Accuracy Evaluation]

Page 9: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

Object Tracking: A Survey

Alper Yilmaz, Omar Javed, Mubarak ShahOhio State University, ObjectVideo, Inc., University of Central FloridaJournal ACM Computing Surveys (CSUR) (2006)

Page 10: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Tracking problems]• Abrupt object motion• Changing appearance (object/scene)• Nonrigid object structures• Obj-to-obj, obj-to-scene occlusions• Camera motion

Page 11: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Key steps in video analysis]• Detection of interesting moving objects• Tracking of object from frame to frame• Analysis of object tracks to recognize their behavior

Page 12: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai
Page 13: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Feature Selection For Tracking]• Color– Spectral power distribution of the illuminant– Surface reflectance properties of the object

• Edges– Generate strong changes in image intensities– Less sensitive to illumination changes

Page 14: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Feature Selection For Tracking]• Optical Flow– Translation of each pixel in a region– Feature in motion-based segmentation and tracking applications

• Texture– Quantifies such as smoothness and regularity– Requires a processing step to generate descriptors

Page 15: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai
Page 16: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai
Page 17: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

Video Retrieval using Spatio-Temporal DescriptorsDaniel DeMenthon, David DoermannUniversity of Maryland MULTIMEDIA Proceedings of the eleventh ACM international conference(2003)

Page 18: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Motivation]• Is an advertising spot been cut by a warning about weather conditions?• Want to jump back to most exciting moments (slow-motion replays) in a football game.• Verify a suspicion of unauthorized access by an outsider of the door system

Page 19: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Video Strands]

Page 20: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai
Page 21: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Indexing and retrieval]• Space-time segmentation– Transform the dynamic content of video clips into simple purely geometric patterns

• Dynamic interaction of color regions in videos for the use of pattern recognition techniques• Combining k-nearest neighbors search and voting on the retrieved labels

Page 22: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Indexing and retrieval]• High level of resilience against video clip variability caused by editing and overlays.• Provide good discriminative power in recognition of actions occurring at fixed places in the field of view of fixed surveillance cameras

Page 23: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

Online learning of robust object detectors during unstable trackingZdenek Kalal, Jiri Matas, Krystian MikolajczykUniversity of Surrey, Czech Technical UniversityComputer Vision Workshops(2009)

Page 24: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Investigates the problem of ]• Visual tracking of unknown objects in unconstrained environments• Cope with frame-cuts, fast camera movements• Partial/total object occlusions/dissapearances• ”Long-term” – possibly infinite length• ”Online” – Need no information from the future

Page 25: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Standard tracking approaches]• Static models– Object appearance change is limited and known. Unexpected changes of the object appearance can not be tracked

• Adaptive method–Update the object model during tracking. But incorrect update brings error

Page 26: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Adaptive tracker]• Trajectory is observed by two processes (growing and pruning event)• Both events make errors, the stability of the system is achieved by their cancelation• The learnt detector enables re-initialization of the tracker whenever previously observed appearance reoccurs

Page 27: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Online model]

Page 28: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai
Page 29: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

Face-TLD: Tracking-Learning-Detection applied to faces Zdenek Kalal, Krystian Mikolajczyk, Jiri MatasUniversity of Surrey, Czech Technical UniversityImage Processing (ICIP) 17th IEEE International Conference (2010)

Page 30: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai
Page 31: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai
Page 32: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[ABOLUTE, CHANGE, LOOP strategy]

Page 33: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai
Page 34: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

[Contributions]• Addresses the long term tracking problem• Learning method based on two events that boot straps the object model from a single click• Efficient detector structure enabling real-time learning/classification

Page 35: Topic regards: ◆ Browsing of Search Results ◆ Video Retrieval using Spatio-Temporal ◆ Object Tracking ◆ Face tracking Yuan-Hao Lai

Thank You.