Syllabus
1. Overview of digital image processing (digital image properties, intensity and geometry transformations, pre-processing, edge detection and interest points / areas).
2. Objects in the image, necessity of interpretation, segmentation of objects.
3. 3D computer vision, goals and applications, overview of the subject. Practical optics. Radial distortion.
4. Camera, projection matrix and its decomposition.
5. Geometry for describing a solid body in 3D. Projective transformation.
6. Geometry of one perspective camera. One camera calibration.
7. Geometry of two cameras, epipolar restriction, fundamental matrix. Geometry of multiple cameras.
8. Essential matrix and its decomposition. 7-point algorithm for estimating the fundamental matrix, 5-point algorithm for estimating the essential matrix.
9. The correspondence problem and its solution.
10. Reconstruction of 3D scene from several views. Shape from motion.
11. Other algorithms for 3D reconstruction, Random Sampling Optimization (RANSAC), Bundle Adjustment.
12. Depth sensors (lidar, radar, sonar, structured light cameras)
13. Merging information from different depth sensors.
14. Practical application in a self-driving car.
Annotation
The subject introduces methods of 3D computer vision. First, in three lectures, we summarize the necessary concepts of digital image processing / analysis.
We follow with the description of one (intensity, RGB) camera and related geometry. We will learn how to use two or more cameras to determine the depth in the scene and for 3D reconstruction.
We will add further depth sensory modalities, namely lidars, radars, sonars. We will explain how to combine information from a variety of sensors.
We will practically illustrate the methods in the tasks associated with self-driving cars.