Computer Vision

Course description

Images are everywhere and so intuitive for humans, but require complex analysis of colors, texture and geometry for a computer. Through a mix of lectures and practices this class will introduce computer vision and the key algorithms to extract semantic information, objects or structure. The course will cover introduction to pixel-level representations to segment textures, shadows, skins or to detect simple objects; model fitting techniques to extract geometrical information in the scene. Using spatial information and learning techniques, we’ll build higher level representation used for scene semantic labeling, or human pose/skeleton estimation which is required for interactive use. Finally, we’ll introduce time and motion processing to estimate the scene structure such as its geometry and dynamics.

Objectives

The class will require knowledge of signal processing and coding skills (Python). The students shall acquire the following knowledge:

  • Introduction to computer vision
  • Texture segmentation (colors, light, texture, objects)
  • Clustering and model fitting (geometry)
  • Estimation of image semantic (pixel-wise labeling)
  • Silhouette and Skeleton extraction (pose estimation)
  • Time and Structure from Motion (tracking, reconstruction)

Academic Instructor

Raoul de Charette | INRIA