Monocular Group Meeting People Localization and Other Projects By Gerald Dalley MIT AI Lab Vision Retreat 14 January 2003.

Monocular Group Meeting Monocular Group Meeting People Localization and People Localization and

Other ProjectsOther Projects

By Gerald DalleyBy Gerald Dalley

MIT AI Lab Vision Retreat MIT AI Lab Vision Retreat

14 January 200314 January 2003

OverviewOverview

Who Am I?Who Am I? 11stst year Ph.D. student in Eric Grimson’s year Ph.D. student in Eric Grimson’s

activity groupactivity group Past and Present Projects…Past and Present Projects…

People LocalizationPeople Localization System designSystem design ChallengesChallenges

Who Am I: Past and Present Projects:Who Am I: Past and Present Projects:

Range Image Registration (B.S. @ OSU)Range Image Registration (B.S. @ OSU)

“earthmover detected”

Who Am I: Past and Present Projects:Who Am I: Past and Present Projects:

Vehicle Recognition From Cluttered Vehicle Recognition From Cluttered Range Images (M.S. @ OSU)Range Images (M.S. @ OSU)

Who Am I: Past and Present Projects:Who Am I: Past and Present Projects:

Equations and Figures: Matlab, PowerPoint, Equations and Figures: Matlab, PowerPoint, Word, LATEX, and More Word, LATEX, and More (current)(current)

Making equations…Making equations…Matlab figure tricksMatlab figure tricksEmbedding LaTeX in PowerPointEmbedding LaTeX in PowerPoint

Tell me about your problems (and/or solutions)Tell me about your problems (and/or solutions)

Who Am I: Past and Present Projects:Who Am I: Past and Present Projects:

HID Consistency Constraints (current)HID Consistency Constraints (current)

Who Am I: Past and Present Projects:Who Am I: Past and Present Projects:

People Localization (Microsoft + MIT)People Localization (Microsoft + MIT)

(267,195,180), 270o

(282,182,142), 292o

(219, 26,149), 270o

(176,188,139), 270o

People Localization: People Localization:

System DesignSystem Design

Video Edges

Motion

Color

Camera Calibration

Head & Shoulders Model

Template Generator

Outline

Comparison

ChallengesChallenges

Inherent challengesInherent challenges No stereoNo stereo Widely-varying occlusion levelsWidely-varying occlusion levels Low contrast between walls and peopleLow contrast between walls and people ScaleScale

Implementation challengesImplementation challenges Chamfer, Hausdorff, Chamfer, Hausdorff, etc.etc. distances have not distances have not

been good discriminators yetbeen good discriminators yet