Content
• basic problem of 3D vision
• comparison of humans and machines
• geometry of 3D objects
• classic geometric methods of 3D computer vision
• deep learning methods of 3D computer vision
• experimental techniques of measuring 3D object properties
• evaluation methods of comparing predictions to direct measurements
• applications, for example autonomous cars
Learning objectives
The problem is how to determine 3D properties of physical objects (shape, position, attitude) from the 2D images of a camera. Presently there are two main classes of solutions: geometry-based computer vision methods, and systems based on neural networks (deep learning) that require training with a large dataset of examples. Both approaches have their strengths and weaknesses and practical vision systems often combine both of them.
• The student has a clear overview of the most important 3D object detection techniques.
• The student can choose the best 3D object detection technique for a given problem.
• The student can evaluate the results of any 3D object detection technique.
Prerequisites
Experience with Python is a necessity. Some basic mathematical knowledge is also required,:
geometry and algebra. Experience with calculus is a benefit but not a prerequisite.