Talking Glasses - [Second Seminar]

Slide 6:

Project Review Objective Problem Definition Project Architecture Methodology Depth Estimation(stereo vision) Object Representation(Codebook) Learning Classification Our Progress Results What’s next ? References

Slide 8:

Build Object Recognition system for outdoors objects

Slide 9:

Active area of Research

Slide 10:

Help Blind people

Slide 12:

Classifier Classification Feature Extraction Post Processing Data Acquisition (Stereo Camera) Depth Estimation Segmentation Narrator (Text To Speech Engine) Codebook

Slide 14:

Depth Estimation

Slide 15:

Stereo Pipeline Epipolar Rectification Stereo Matching Depth via Triangulation Rectified Left Image Rectified Right Image Disparity Map 3D Scene Reconstruction Left Image Right Image

Slide 16:

Local Methods 1-A Naive Stereo Algorithm

Slide 17:

Local Methods 2 - Window-Based Matching 17 (a) Left image (b) Right image Find point of maximum correspondence Compare color values within search windows

Slide 18:

Local Methods Window size = 3x3 pixels Window size = 21x21 pixels 2 - Window-Based Matching

Slide 19:

Global Methods Stereo as an Energy Minimization Problem

Slide 20:

Bayesian Approach

Slide 21:

Bayesian Approach

Slide 22:

Bayesian Approach

Slide 23:

Object Representation(Codebook)

Slide 25:

Step1: Feature Extraction Detect Interest Points Extract Patches. Describe Patches Common detectors and Descriptors: SIFT SURF

Slide 26:

SURF vs. SIFT

Slide 27:

Step2: Clustering K-means Clustering. RNN Clustering.

Slide 28:

RNN vs. K-means

Codebook Module:

Codebook Module Training Data Feature Extraction Interest Point Detector Descriptor Clustering Codebook

Learning:

Learning

Learn Spatial Relation between features:

Learn Spatial Relation between features

Learn Spatial Relation between features:

Learn Spatial Relation between features

Learning Module:

Implicit Shape Model (ISM) Learning Module Training Data + Reference Segmentation Feature Extraction Codebook M atching Spatial Probability

Classification:

Classification

Voting Approach :

Voting Approach

Classification Module:

Implicit Shape Model (ISM) Classification Module Testing Data Feature Extraction Matching Voting Space

Support Vector Machine(SVM):

Support Vector Machine(SVM)

Over-fitting Problem:

Over-fitting Problem

Decision Boundary :

Decision Boundary

Pairwise SVM:

Pairwise SVM

Slide 44:

H . Bay, T. Tuytelaars , and L. Van Gool . Surf: Speeded up robust features. European Conference on Computer Vision, 1:404-417, 2006 . Christopher Evans. Notes on the OpenSURF Library, January 18, 2009 B . Leibe , A. Leonardis , and B. Schiele . Robust object detection with interleaved categorization and segmentation. International Journal of Computer Vision, 77(1-3):259–289, 2008b . Visual Vocabularies for Category-Level Object Recognition, Ph.D. Thesis, R. J. López-Sastre.2009 S . A. Nene and S. K. Nayar . A simple algorithm for nearest neighbor search in high dimensions. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(9):989–1003, 1997 .