Computer Vision No. 1 What is the Computer Vision?
Instructor u Katsushi Ikeuchi u Pointers: Komaba Meguro-ku
Evaluation u attendance 50% u report 50%
Schedule u Shape-from-X –Analysis of line-drawing –Shape-from-shading –Binocular stereo u Interpretation –Interpolation –Representation u Special topics –Modeling from reality
Katsu Ikeuchi U. Tokyo Human visual system MIT AI Shape-from-shading ETL Object recognition CMU Assembly plan from observation Modeling from reality U. Tokyo Virtual heritage
Demonstration Videos
Photometric Stereo (1980) u Brightness difference -> 3D shape u 3D shape -> 3D Pose determination u 3DPose -> Grasping
Bin Picking
Assembly Plan from Observation (1990)
Recent Result Assembly plan from observation
Learning Human Dance
Motion Capture Data
Robot Dancing
Modeling Cultural Heritage
Virtual City Probe Info
Virtual City Speed : 10km/h Vehicle Pedestrian Near Yoyogi park
Computer Vision (CV) u To make a computer to recognize the 3D world as we do u To generate 3D representations from 2D images
CV and related areas Image Understanding (AI) Pattern Recognition (Mathematical theories) Image Processing (Signal processing)
CV and related areas Image Understanding (AI) Pattern Recognition (Mathematical theories) Image Processing (Signal processing)
Image Processing To get better images: 2D-to-2D
CV and related areas Image Understanding (AI) Pattern Recognition (Mathematical theories) Image Processing (Signal processing)
Pattern Recognition Decision making: mathematical theories
CV and related areas Image Understanding (AI) Pattern Recognition (Mathematical theories) Image Processing (Signal processing)
Image Understanding Scene description
Why difficult ? u A lot of data u Ambiguity –Projection of a 3D world to a 2D image u Many factors to influence the image –Illumination condition –Object shape –Camera characteristics
Image Foggy golden triangle in Pittsburgh
But …
A lot of data u Landsat image –1scene: 3300 x 2300 x 4 = bytes –200 scenes/ day u Color TV image –512 x 512 x 3 x 30 = bytes/sec
Why difficult ? u A lot of data u Ambiguity –Projection of a 3D world to a 2D image u Many factors to influence the image –Illumination condition –Object shape –Camera characteristics
Illusion due to the projection
Why difficult ? u A lot of data u Ambiguity –Projection of a 3D world to a 2D image u Many factors to influence to the image –Illumination condition –Object shape –Camera characteristics
Image u A image is a matrix of pixels u Each pixel –brightness –Color –Distance
Inside and Outside (Gestalt)
Common sense u To formulate the common sense → research topics
Current issues u A lot of data –Computational sensor –Vision board u Ambiguity –Projective geometry –constraints u Many factors –Physics-based vision
Application areas
What is Computer Vision? u Vision is … an information processing task that constructs efficient symbolic descriptions of the world from images. (Marr) u Vision is … inverse graphics. u Vision is … looks easy, but is difficult. Vision is … difficult, but is fun. (Kanade) u Vision is an engineering science to create an alternative of human visual systems on computers ( Ikeuchi )
References u Journals –Inter. J. Computer Vision –IEEE Trans. Pattern Analysis and Machine Intelligence –IEICE D-2 –IPSJ Trans CVIM u International conferences –Inter. Conf. Computer Vision (ICCV) –Computer Vision and Pattern Recognition (CVPR) –Asian Conf. Computer Vision (ACCV) u Special interest groups –IPSJ CVIM –IEICE PRMU
Schedule (April-May) 4/12 Introduction 4/19 Line drawing 4/26 Perspective projection 5/3 Holiday 5/10 Shape from Shading 5/17 Color Dr. Miyazaki 5/24 Stereo#1 5/31 Stereo#2 Dr. Vanno and Dr. Ogawara
Schedule (June-July) 6/7 Motion analysis 6/14 No class 6/21 EPI, IBR & MBR Dr. Ono 6/28 Interpolation 7/5 Object representation#1 Dr.Takamatsu 7/12 Object representation#2