Oleh Tretiak © Computer Vision Lecture 1: Introduction
Oleh Tretiak © Introduction: Administrative Instructor: –Oleh Tretiak –Course web site: –Office: Lviv Polytechnic, Building 5, room 801 –Office hours: Thursdays 12-2 Textbook: Дэвид Форсайт, Жан Понс, (David Forsythe, Jean Ponce) Компьютерное зрение – современний подход, Вильямс (Москва, Санкт- Петербург, Киев), 2004 Textbook web site:
Oleh Tretiak © Syllabus (see course web site for more details) 1.Introduction, camera model 2.Linear Filters 3.Edge detection and texture 4.Multi-image and stereo 5.Segmentation and structural operations 6.Segmentation and probabilistic methods 7.Recognition through template matching 8.Classification and evaluation
Oleh Tretiak © Artificial Intelligence and Computer Vision Computer Vision: production of information about the physical world from optical sensors Type of information –Non-contact sensing –Interpreting symbol, e. g. optical character recognition –Information about three-dimensional objects (distance, obstacles) Computer vision is part of the functioning of autonomous agents
Oleh Tretiak © Computer Vision and Related Areas Image Processing: Formation and enhancement of images. For example, Computer Tomography Machine Vision: Automated sensing and classification in manufacturing Robot Vision: Control of vehicles and manufacturing devices Computer Graphics: Many computer and mathematical tools are shared with Computer Vision
Oleh Tretiak © Classes of Vision Tasks Reflexive –Full task consists of sensing and response. Sensor that actuates a supermarket checkout belt drive Multi-level –Reflexive task guided by dynamical process Optical character recognition The dynamical process may be guided by an explicit model of the object being analyzed
Oleh Tretiak © Conceptual Structure of Computer Vision Image-object relation –Physics and optics of cameras –Photometry –Color Early vision (first layer) –One image Edge detection Texture –Multiple images Stereo vision for depth information Shape from motion
Oleh Tretiak © Conceptual Structure Mid-level vision (second layer) –Segmentation Find objects in image by grouping similar areas Find objects in sequence of images by finding regions that move together
Oleh Tretiak © Structure of Vision High level vision (third layer) –Geometry: Model used to find known objects Model used to find change of shape due to motion –Probability: Classifiers to find objects Templates
Oleh Tretiak © Lecture Outline Cameras and perspective projection (Section 1.1 in the textbook)
Oleh Tretiak © Pinhole Camera
Oleh Tretiak © Distant Objects Have Smaller Images
Oleh Tretiak © Parallel Lines Meet at Infinity
Oleh Tretiak © Equations of Projection x’ = fx/z y’ = fy/z z’ = f
Oleh Tretiak © Common Approximations Projection equations are nonlinear Weak perspective: –Magnification is constant over a ‘thin’ object Orthographic: –x’ = x, y’ = y Affine –x’ = ax + by + cz + d –y’ = ex + fy + gz + h Accounts for object rotation, shift Valid for small z changes (locally affine)
Oleh Tretiak © Real Cameras Lenses are used to collect more light –Pinhole camera admits very little light Lenses introduce distortions (geometric distortion, defocusing) Images are recorded with electronic sensors –Obtain rectangular arrays of numbers