Computer Vision (CV)

• Official homepage (with schedule)

Description

The objective of this computer vision module is to provide students with fundamental knowledge in the field of 3D scene analysis and in particular of the underlying 3D geometry. The geometric aspects of image processing and computer vision are presented. From a practical point of view, these are the necessary basics for understanding 3D perception. It is also an important link with image synthesis and animation courses.

Keywords

Computer vision, motion analysis, geometry.

Prerequisites

Basic knowledge of linear algebra, optimization

Content

• Geometric Aspect of Vision
  • Camera model, calibration (SVD)
  • Stereovision and other reconstruction methods
  • Multi-view geometry, SFM, epipolar geometry estimation, homography, image transfer
  • An application example: Motion capture technologies
  • Camera localization, pose estimation, PnP, SFM (real-time and offline bundle adjustment, Matchmoving), SLAM
• Motion estimation
  • Classical methods: Lucas-Kanade, Global approach with regularization (Horn & Shunk), large displacement, multi-resolution, discontinuity preservation (robust estimation).
• Tracking
  • Small dimension: Shi-Tomasi-Kanade, Kalman, EKF
  • Larger size system: ensemble Kalman filter, data assimilation, optimal control (velocity field estimation in meteorological imagery)

Acquired skills

The knowledge necessary to create special effects, to design autonomous vehicles, to analyze images in an environmental context.

Professors

Eric Marchand (in charge), Etienne Mémin