Video AlignmentIntroductionAlgorithm [Caspi and Irani, 2002]Algorithm (continued)ExtensionResults - Finding Transformations [Bergen, et. al., 1992]Video AlignmentVideo AlignmentPresented byOmer ShakilPresented byOmer ShakilIntroductionIntroductionProblem StatementAlignment TechniquesDirect-Based [Caspi and Irani, 2000 and 2002]Feature-Based [Stein, 1998]Best Frame Pair Search [Sand and Teller, 2004]Non-Overlapping SequencesCoherent Temporal Behavior [Caspi and Irani, 2001 and 2002]Problem StatementAlignment TechniquesDirect-Based [Caspi and Irani, 2000 and 2002]Feature-Based [Stein, 1998]Best Frame Pair Search [Sand and Teller, 2004]Non-Overlapping SequencesCoherent Temporal Behavior [Caspi and Irani, 2001 and 2002]Algorithm [Caspi and Irani, 2002]Algorithm [Caspi and Irani, 2002]Video Sequences S and S’Inter-Camera HomographyInter-Frame Transformation Ti between Ii and Ii+1Derive orHenceLeast Square MinimizationVideo Sequences S and S’Inter-Camera HomographyInter-Frame Transformation Ti between Ii and Ii+1Derive orHenceLeast Square Minimization1' 'i i ip T p+@1i i ip T p+@1 1'i ip Hp+ +@'i ip Hp@1'i iT HT H-@1'i i iT s HT H-@( ' ) ( )i i ieig T s eig T@[ ]1 2 3( ) , ,Teig A l l l=' 0i i is HT T H- =Mi is a 9 x 9 matrix defined by Ti, T’i and siCombining all the constraintsH = Eigen-vector of ATA corresponding to the smallest eigen-value.Mi is a 9 x 9 matrix defined by Ti, T’i and siCombining all the constraintsH = Eigen-vector of ATA corresponding to the smallest eigen-value.Algorithm (continued)Algorithm (continued)[ ]11 12 13 21 22 23 31 32 33Th H H H H H H H H H=r0iM h =r r0Ah =r r[ ]1 2...TnA M M M=Learning PhaseRecover Inter-Camera HomographyFree Camera MovementFrame RegistrationReference Frame from the Learning PhaseFuture Work3D ScenesSynchronizationLearning PhaseRecover Inter-Camera HomographyFree Camera MovementFrame RegistrationReference Frame from the Learning PhaseFuture Work3D ScenesSynchronizationExtensionExtensionInter Frame TransformationsInputSequencesInter CameraHomographyFrameRegistrationAlignedSequencesResults - Finding Transformations [Bergen, et. al., 1992]Results - Finding Transformations [Bergen, et. al., 1992]1.2 0 100 1.2 100 0 1-� �� �-� �� �� �1 0 100 1 200 0 1� �� �� �� �� �1.2000 0.0000 9.99920.0000 1.1999 9.99490 0 1-� �� �-� �� �� �Scaling Translation RotationActualRecoveredImages1.0000 0.0000 10.00000.0000 1.0000 20.00000 0 1� �� �� �� �� �0.9397 0.3420 200.3420 0.9397 300 0 1-� �� �-� �� �� �0.9397 0.3420 20.00180.3420 0.9396 29.99310 0 1-� �� �-� �� ��
View Full Document