16
Motion Map: Image-based Retri eval and Segmentation of Motion Da ta EG SCA ’04 EG SCA ’04 Speaker: Alvin Speaker: Alvin Date: 11/29/2004 Date: 11/29/2004

Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

  • View
    216

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Motion Map: Image-based Retrievaland Segmentation of Motion Data

EG SCA ’04EG SCA ’04

Speaker: AlvinSpeaker: Alvin

Date: 11/29/2004Date: 11/29/2004

Page 2: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

22

OutlineOutline

IntroductionIntroduction FrameworkFramework ResultsResults Evaluation FormEvaluation Form ConclusionsConclusions Future WorksFuture Works

Page 3: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

33

IntroductionIntroduction

Provide Provide GUIGUI for retrieving motion data. for retrieving motion data. Using Self-organizing map. (Using Self-organizing map. (SOMSOM)) Only need to specify starting and ending pOnly need to specify starting and ending p

ostures.ostures.

Page 4: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

44

SOMSOM

A type of unsupervised learning.A type of unsupervised learning. Usually 1D or 2DUsually 1D or 2D A A mappingmapping that that preserves neighborhood represerves neighborhood re

lationslations.. All nodes in input layer are connected to alAll nodes in input layer are connected to al

l nodes in output layer.l nodes in output layer. Often used in information visualization.Often used in information visualization. DetailDetail

Page 5: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

55

SOMSOM

Extracted from 55,114 frames of 51 data files. It took about 17 minutes on P4 2.4G for 21 * 21 nodes.

Page 6: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

66

ClusteringClustering

Divides regions by Divides regions by detecting bordersdetecting borders The average difference against 4 The average difference against 4

neighborsneighbors

Create vertical border ifCreate vertical border if

LabelingLabeling

Page 7: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

77

Posture IconsPosture Icons

From the node that is nearest to the center From the node that is nearest to the center of each clustered region.of each clustered region.

Page 8: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

88

TrajectoryTrajectory

Each motion can be represented as a Each motion can be represented as a trajectory.trajectory.

The walking motion:The walking motion:

Page 9: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

99

Virtual NodeVirtual Node

Increase the resolution Increase the resolution with small computation with small computation cost.cost.

Can be Can be preprocessedpreprocessed f for great detail with the or great detail with the cost of storage.cost of storage.

Page 10: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

1010

RetrievalRetrieval

Page 11: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

1111

ResultsResults

Page 12: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

1212

ResultsResults

Page 13: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

1313

Evaluation FormEvaluation Form

論文簡報部份論文簡報部份 完整性介紹完整性介紹 (4)(4) 系統性介紹系統性介紹 (4)(4) 表達能力表達能力 (3)(3) 投影片製作投影片製作 (3)(3)

論文審閱部分論文審閱部分 瞭解論文內容瞭解論文內容 (4)(4) 結果正確性與完整性 結果正確性與完整性 (4)(4) 原創性與重要性原創性與重要性 (4)(4) 讀後啟發與應用:讀後啟發與應用:

The SOM can be used either for clustering or classification in my research.Besides, because the GUI is very friendly, maybe we can exploit it in our 套路 editor. Additionally, we may use the concept of LOD into the SOM.

Page 14: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

1414

ConclusionsConclusions

Contributions:Contributions: AutomaticallyAutomatically Easily RetrieveEasily Retrieve Display motion as a trajectoryDisplay motion as a trajectory

Defects:Defects: Can’t distinguish different performersCan’t distinguish different performers Can’t reflect the dynamical featureCan’t reflect the dynamical feature

Page 15: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

1515

Future WorksFuture Works

Analyzing minute difference.Analyzing minute difference. Zooming in the motion trajectories.Zooming in the motion trajectories.

Interactive data editing.Interactive data editing. A intuitive tool for evaluating motion graphs.A intuitive tool for evaluating motion graphs. Motion blending by drawing a path.Motion blending by drawing a path.

Page 16: Motion Map: Image-based Retrieval and Segmentation of Motion Data EG SCA ’04 Speaker: Alvin Date: 11/29/2004

Alivn/GAME Lab./CSIE/NDHUAlivn/GAME Lab./CSIE/NDHU Motion Map: Image-based Retrieval and SegmentatiMotion Map: Image-based Retrieval and Segmentation of Motion Dataon of Motion Data

1616

SOMSOM

Nj denotes the total number of joints

t = 0,1,2,… denotes the number of iterations.