CS 188: Artificial Intelligence Perceptual and Sensory Augmented - PDF document

4/26/12 ¡ Rough evolution of focus in recognition research CS 188: Artificial Intelligence Perceptual and Sensory Augmented Computing Visual Object Recognition Tutorial Lecture 24: Computer Vision Pieter Abbeel – UC Berkeley 1980s 1990s to early 2000s 2000-2010… Slides adapted from Trevor Darrell (and his sources) Inputs/outputs/assumptions • What is the goal ? – Say yes/no as to whether an object present in image And/or: – Determine pose of an object, e.g. for robot to grasp – Categorize all objects Scanning windows … – Forced choice from pool of categories – Bounding box on object – Full segmentation – Build a model of an object category Detection via classification: Main idea Detection via classification: Main idea Basic component: a binary classifier If object may be in a cluttered scene, slide a window around looking for it. Perceptual and Sensory Augmented Computing Perceptual and Sensory Augmented Computing Visual Object Recognition Tutorial Visual Object Recognition Tutorial Visual Object Recognition Tutorial Visual Object Recognition Tutorial Car/non-car Car/non-car Classifier Classifier No, not a car. Yes, car. K. Grauman, B. Leibe K. Grauman, B. Leibe 1 ¡

4/26/12 ¡ Detection via classification: Main idea Detection via classification: Main idea Fleshing out this • Consider all subwindows in an image pipeline a bit more, Ø Sample at multiple scales and positions (and orientations) we need to: Perceptual and Sensory Augmented Computing Perceptual and Sensory Augmented Computing 1. Obtain training data • Make a decision per window: 2. Define features 3. Define classifier Ø “Does this contain object category X or not?” Visual Object Recognition Tutorial Visual Object Recognition Tutorial Visual Object Recognition Tutorial Visual Object Recognition Tutorial Training examples Car/non-car Classifier Feature extraction 8 K. Grauman, B. Leibe K. Grauman, B. Leibe Feature extraction: Eigenfaces: global appearance description global appearance An early appearance-based approach to face recognition Feature extraction Generate low- Perceptual and Sensory Augmented Computing Perceptual and Sensory Augmented Computing dimensional representation of appearance Mean with a linear Visual Object Recognition Tutorial Visual Object Recognition Tutorial Visual Object Recognition Tutorial Visual Object Recognition Tutorial Eigenvectors computed from subspace. Training images covariance matrix Project new Simple holistic descriptions of image content ... images to “face Ø grayscale / color histogram ≈ + + space”. + + Ø vector of pixel intensities Mean Recognition via nearest neighbors in face space Turk & Pentland, 1991 K. Grauman, B. Leibe K. Grauman, B. Leibe Gradient-based representations Feature extraction: global appearance • Pixel-based representations sensitive to small shifts • Consider edges, contours, and (oriented) intensity gradients Perceptual and Sensory Augmented Computing Perceptual and Sensory Augmented Computing Visual Object Recognition Tutorial Visual Object Recognition Tutorial Visual Object Recognition Tutorial Visual Object Recognition Tutorial • Color or grayscale-based appearance description can be sensitive to illumination and intra-class appearance variation Cartoon example: an albino koala K. Grauman, B. Leibe K. Grauman, B. Leibe 2 ¡

4/26/12 ¡ Gradient-based representations: Histograms of oriented gradients (HoG) HOG Perceptual and Sensory Augmented Computing Visual Object Recognition Tutorial Visual Object Recognition Tutorial (one of the most widely used features) Map each grid cell in the input window to a histogram counting the gradients per orientation. Code available: http://pascal.inrialpes.fr/ soft/olt/ Dalal & Triggs, CVPR 2005 K. Grauman, B. Leibe Slide credit: Dalal, Triggs, P . Barnum Slide credit: Dalal, Triggs, P . Barnum • Histogram of gradient orientations centered -Orientation -Position diagonal uncentered – Weighted by magnitude cubic- Sobel corrected Slide credit: Dalal, Triggs, P . Barnum Slide credit: Dalal, Triggs, P . Barnum 3 ¡

4/26/12 ¡ Slide credit: Dalal, Triggs, P . Barnum Slide credit: Dalal, Triggs, P . Barnum Slide credit: Dalal, Triggs, P . Barnum Slide credit: Dalal, Triggs, P . Barnum 4 ¡