72530 - Advanced Image Analysis and Computer Vision Techniques

Learning outcomes

Course contents

Introduction

Definition of Computer Vision. Definition of image processing and analysis system. Some applications: smart video surveillance (video analytics); people tracking for behaviour analysis; automatic event detection and analysis in traffic monitoring; automatic vehicle guidance, Unmanned Aerial Vehicles (UAV) and satellite attitude control; medical (CT, MR, …) and biomedical (microscopic cell cultures and histological samples) image analysis; automatic measurements and quality control in machine vision; robotics and 3D scene reconstruction.

Image formation

Image types: photographic, thermal, radiographic images. Gray levels and colour channels. Image formation process and system and digitalization. Optical device: lens, aberration, focus, iris. Radiometry: radiance and irradiance. Vignetting detection and correction. Quantization, radiometric resolution and dynamic range. Camera Response Function (RF). Image acquisition and sampling process. Spatial resolution and Point Spread Function (PSF). Images from non optical devices.

Camera model and geometric transform

Shot geometry: sensor position and perspectives. World, camera, image coordinate systems. Geometric operations and transforms: interpolation, scaling, rotation, translation, affine, perspective deformation. Camera parameters and geometric calibration: intrinsics and extrinsics. Weak and strong perspective model. Homographies and perspective matrix. Patterns and calibration techniques. Laboratory: camera calibration.

Scene segmentation

Histogram definition, property and shape. Cumulative histogram. Histogram's use in image processing and analysis. Segmentation of stationary scenes. Outdoor and indoor segmentation: shadow and reflection. Gradient-based segmentation and Hough transform. Object segmentation: fixed and adaptive thresholding. Color based segmentation, split and merge, watershed, non- parametric k-means. Labelling. Applications: computer graphics, automatic object detection, automatic light condition stabilization.

Pattern recognition and image analysis with multidimensional data

Recall of probability and statistics for data analysis. Probability density and distribution function. 2D Gaussian function. Decision theory and multivariate data classification. Automatic image feature extraction. Photometrical, geometrical, statistical features. Principle of texture analysis and moments. Multidimensional features and dimensionality reduction techniques. Supervised classification and clustering. Proximity measures. Bayesian classification, MAP and MLE. Automatic object recognition. Applications: automatic object detection and recognition, machine vision.

Video and image sequence analysis

Acquisition technologies, video format and codecs. Automatic motion detection and analysis: frame differencing, background modelling and updating. Segmentation of moving objects. Random sampling algorithms. Optical flow and Lucas-Kanade corner extraction and tracking. Coarse-to-fine (pyramidal) approach. Parametric motion modelling. Feature extraction and tracking: classical and modern trackers and their applications. Automatic features and object tracking. Semantic labelling. Applications: security, automatic event detection, quality control, machine vision. Image stabilization.

Multiple view geometry and scene reconstruction

Image registration techniques and pattern matching. Image registration: local and global methods. Image mosaicing: frame-to-frame and frame-to-mosaic registration. Dead reckoning and looping path problem. Image warping, interpolation and blending. 3D stereopsis: principles and techniques. Feature based and dense stereoscopy. Epipolar geometry. Stereo system calibration and image rectification. Attitude and pose recovery of moving cameras. 3D image formation from multiple 2D images. 3D scene reconstruction from single monocular camera. Applications: computer graphics, image metrology and automatic static and dynamic object measurements, automatic vehicle guidance and control. Panorama.

Real world computer vision systems: case studies

Experimental characterization of camera RF and PSF. High dynamic range imaging. Calibration of a stereo couple. Real time image mosaicing. Detection and tracking of moving objects. 3D scene reconstruction from multiple views.

Readings/Bibliography

• R. Gonzales, R. Woods: “Digital Image Processing”, Second Edition, Prentice-Hall, New-Jersey, USA, 2002
• R. I. Hartley, A. Zisserman: “Multiple View Geometry in Computer Vision”, Second Edition, Cambridge University Press, 2004
• Richard O. Duda, Peter E. Hart, David G. Stork: “Pattern Classification”, Second Edition, Wiley Interscience, New York, 2001
• CVonline: Vision Related Books including Online Books and Book Support Sites (http://homepages.inf.ed.ac.uk/rbf/CVonline/books.htm)

Teaching methods

Classroom lessons and practice in lab. Each topic will be treated jointly with significant case studies developed in lab to highlight its meaningful applications. In order to make the students aware of the different topics, many homework exercises will be proposed and publicly corrected afterwards in labs.

Assessment methods

The students will be evaluated on the basis of a computer vision project, besides through a classroom practice and oral examination.

Teaching tools

In the teaching material section all the slides shown in class are available for download as well as the software tools for practice in lab.

Links to further information

http://cvg.deis.unibo.it

Office hours

See the website of Alessandro Bevilacqua