Computer Science Department Pacific University Artificial Intelligence -- Computer Vision

Why Computer Vision? Vision, communication, & action Pacific University

Why study Computer Vision? F Images and video are everywhere F Fast-growing collection of useful applications F matching and modifying images from digital cameras F film special effects and post-processing F building representations of the 3D world from pictures F medical imaging, household robots, security, traffic control, cell phone location, face finding, video game interfaces,... F Various deep and attractive scientific mysteries F what can we know from an image? F how does object recognition work? F Greater understanding of human vision and the brain F about 25% of the human brain is devoted to vision F Images and video are everywhere F Fast-growing collection of useful applications F matching and modifying images from digital cameras F film special effects and post-processing F building representations of the 3D world from pictures F medical imaging, household robots, security, traffic control, cell phone location, face finding, video game interfaces,... F Various deep and attractive scientific mysteries F what can we know from an image? F how does object recognition work? F Greater understanding of human vision and the brain F about 25% of the human brain is devoted to vision Pacific University

Why is Vision Interesting? F Psychology F ~ 50% of cerebral cortex is for vision. F Vision is how we experience the world. F Engineering F Want machines to interact with world. F Digital images are everywhere. F Psychology F ~ 50% of cerebral cortex is for vision. F Vision is how we experience the world. F Engineering F Want machines to interact with world. F Digital images are everywhere.

Vision is inferential: Illumination Pacific University

Applications of Computer Vision: Texture generation Input image Pattern Repeated New texture generated from input Simple repetition Pacific University

Application: Football first-down line Requires (1) accurate camera registration; (2) a model for distinguishing foreground from background Pacific University

Application areas:  Film production (the “ match move ” problem)  Heads-up display for cars  Tourism  Architecture  Training Technical challenges:  Recognition of scene  Accurate sub-pixel 3-D pose  Real-time, low latency Application areas:  Film production (the “ match move ” problem)  Heads-up display for cars  Tourism  Architecture  Training Technical challenges:  Recognition of scene  Accurate sub-pixel 3-D pose  Real-time, low latency Application: Augmented Reality Pacific University

Application: Medical augmented Reality Visually guided surgery: recognition and registration Pacific University

Application: Automobile navigation Lane departure warning Pedestrian detection Mobileye (see mobileye.com) Other applications: intelligent cruise control, lane change assist, collision mitigation Systems already used in trucks and high-end cars Pacific University

Tracking (Comaniciu and Meer)

Understanding Action

Tracking and Understanding (

Tracking

Part I: The Physics of Imaging F How images are formed F Cameras F What a camera does F How to tell where the camera was (pose) F Light F How to measure light F What light does at surfaces F How the brightness values we see in cameras are determined F How images are formed F Cameras F What a camera does F How to tell where the camera was (pose) F Light F How to measure light F What light does at surfaces F How the brightness values we see in cameras are determined Pacific University

Part II: Early Vision in One Image F Representing local properties of the image F For three reasons F Sharp changes are important in practice -- find “edges” F We wish to establish correspondence between points in different images, so we need to describe the neighborhood of the points F Representing texture by giving some statistics of the different kinds of small patch present in the texture. F Tigers have lots of bars, few spots F Leopards are the other way F Representing local properties of the image F For three reasons F Sharp changes are important in practice -- find “edges” F We wish to establish correspondence between points in different images, so we need to describe the neighborhood of the points F Representing texture by giving some statistics of the different kinds of small patch present in the texture. F Tigers have lots of bars, few spots F Leopards are the other way Pacific University

Part III: Vision in Multiple Images F The geometry of multiple views F Where could it appear in camera 2 (3, etc.) given it was here in 1? F Stereopsis F What we know about the world from having 2 eyes F Structure from motion F What we know about the world from having many eyes F or, more commonly, our eyes moving. F Correspondence F Which points in the images are projections of the same 3D point? F Solve for positions of all cameras and points. F The geometry of multiple views F Where could it appear in camera 2 (3, etc.) given it was here in 1? F Stereopsis F What we know about the world from having 2 eyes F Structure from motion F What we know about the world from having many eyes F or, more commonly, our eyes moving. F Correspondence F Which points in the images are projections of the same 3D point? F Solve for positions of all cameras and points. Pacific University

Part IV: High Level Vision F Model based vision F find the position and orientation of known objects F Using classifiers and probability to recognize objects F Templates and classifiers F how to find objects that look the same from view to view with a classifier F Relations F break up objects into big, simple parts, find the parts with a classifier, and then reason about the relationships between the parts to find the object F Model based vision F find the position and orientation of known objects F Using classifiers and probability to recognize objects F Templates and classifiers F how to find objects that look the same from view to view with a classifier F Relations F break up objects into big, simple parts, find the parts with a classifier, and then reason about the relationships between the parts to find the object Pacific University

Pacific University

Pacific University

Object and Scene Recognition F Definition: Identify objects or scenes and determine their pose and model parameters F Applications F Industrial automation and inspection F Mobile robots, toys, user interfaces F Location recognition F Digital camera panoramas F 3D scene modeling F Definition: Identify objects or scenes and determine their pose and model parameters F Applications F Industrial automation and inspection F Mobile robots, toys, user interfaces F Location recognition F Digital camera panoramas F 3D scene modeling Pacific University

Invariant Local Features F Image content is transformed into local feature coordinates that are invariant to translation, rotation, scale, and other imaging parameters SIFT Features Pacific University

Examples of view interpolation Pacific University

Recognition using View Interpolation Pacific University