70
Topics in Vision and Graphics Richard Hartley

Topics in Vision and Graphicsusers.cecs.anu.edu.au/~hongdong/Hartley.pdf · 2007. 9. 11. · 3D surface of that object (CAD model). ... E-med project – Collaboration with Lions

  • Upload
    others

  • View
    3

  • Download
    0

Embed Size (px)

Citation preview

  • Topics in Vision and GraphicsRichard Hartley

  • I am not going to talk about this stuff …

  • Sensors:• FaceLab• CeDAR active head• 4 x Sony NTSC cameras• 6 DOF INS• Tail shaft encoder• Steering potentiometer• Strain gauges on steering wheel

    Actuation:• Throttle through cruise control• Steering drive motor• Brake linear actuator

    Experimental Vehicle

  • Speed Sign Detection& Recognition

    High reliability detectors for speed and traffic signs.

  • Example of hyperspectral imagery in the detection of fruit damage.

    Image of an imaged cylinder, mapped onto a 3D surface of that object (CAD model).

    Hyperspectral and IR imagery

    Spectral Imagery

  • Automatic Anatomical Structure ExtractionAnalysis of MR brain images

  • Specularity detection

    a. Input b. Normalized image

    c. Diffuse component

    c. Specular Component

    Color Constancy

    Reflection Separation

    Chief Researcher: Robby Tan

  • New View Synthesis

    •Generation of new-viewpoint images from sets of images or Video

    •Possible application in making 3D video from 2D.

    Key researcher: Hongdong Li

  • Radial Distortion Correction

    Improved methods of detecting and correcting radial distortion

    Key researchers: Hongdong Li, Richard Hartley

  • E-med project – Collaboration with Lions Eye Institute

    • Focus on ophthalmologicalapplications– Diabetic retinopathy– Glaucoma detection through

    stereo imaging– Tear film detection.

    Key researcher: Richard Hartley (+ grad student)

  • Ubiquitous Flying Insect Robot

    Design and use of a small mobile robot and investigation of insect-based vision

    Key researchers: Nick Barnes, M. Srinivasan

  • Background Subtraction and Applications

    But I am going to talk about this …

  • Original Video

    Our original algorithm

  • Synopsis Video

  • More difficult example

    Shyjan Algorithms

    Show cook fountain video

  • Synopsis Video

  • Synopsis with Overlay

  • Indoor Sequence

  • Tracking People

  • Show Soccer Video

  • Applications for Surveillance

  • Measuring dimensions and trajectory

    •The estimated heights are: •Tiberio: 1.79m•Peter: 1.89m•Richard: 1.95m

    Trajectory

  • Robust to change in scale and view point

    Recognition of People in Videos

  • Show Soccer TrackingKylie

    How do you do this?

  • Computer Vision in Hollywood

  • Courtesy Oxford Visual Geometry group

  • Courtesy Oxford Visual Geometry group

  • Courtesy Oxford Visual Geometry group

  • What is the mathematics behind this?

    Projective Geometry

  • Felix Klein: Erlangen Program (1872)

    View of geometry as an analysis of what is invariant under different groups of transforms.

    i.e. What is unchanged about the plane when it undergoes a particular type of transformation (motion, distortion, change).

  • Euclidean GeometryThe geometry of rigid transformations (congruence).

    Invariants are:

    • Angles, distances, areas

    • Circles remain circles

    • Straight lines remain lines

  • Concepts of Euclidean Geometry•Circle

    •Length

    •Area

    •Congruent triangles

    •Parallelism

    •Angles

    All these are invariant (unchanged) under rigid transformations.

  • Projective GeometryThe geometry of projective transforms.

    What transformations does a plane undergo when you take an image of it with a pinhole camera?

  • Parallel lines are not preserved

  • Neither are angles, areas, distance-ratios, lengths

  • Circles are projected as ellipses

    Fisheye imageProjective Image

  • • Circles are not preserved under projective transformations.

    • They can be seen as ellipses (including circles), parabolas, hyperbolas, (second degree curves).

    • The right concept in projective geometry is a “conic”.

    • Conics are preserved by projective transformations.

    “Second Order” curves.

  • Any two lines meet in exactly one point

    Line at infinity

  • Projective TransformationAny mapping of the projective plane to itself that preserves lines is called a “projective transformation”, or “collineation”, or “homography”

    Mapping: A mapping from points in the plane to points in the plane.

    Preserves lines: If three points lie on a line, then their corresponding points under the mapping lie on a line.

  • Tracked Points

  • Billboard: Image in an image

  • Building Models from Video

  • Courtesy Marc Pollefeys

  • Courtesy Marc Pollefeys

  • Single View Reconstruction

  • Original Painting

    Synthetic Views

  • Interactive Video

    Interactive video navigation using a game-controller.

  • Navigation is done using a game controller.

  • The End