INF 2064 – Suggestions for the semester project

INF 2064 – Suggestions for the semester project

The time is passing, and we only have eight weeks to finish the course. If you have a clear idea of the work you intend to do for this course, please send me your proposal by e-mail by next Tuesday, 29, and I will reply with my comments. On the other hand, if by next Tuesday, you still do not have a proposal in the following, there is a suggestion for you to achieve a successful closure of the course. The suggestion has several items. Let's see how many of the items you can complete until the last day of classes.

Sugestion: Computer vision tutorial ("é ensinando que se aprende")

One of the best ways to learn the complexity of Computer Vision Algorithms is to explain them in a series of Notebooks Python with synthetic images generated by Computer Graphics. With the syntetic photos we know all about the scene and the camera and we can compare the results. Here are some suggestions:

  1. Use a rendering algorithm to create a picture of a 3D scene with a set of N (say 31) points with known 3D coordinates. Read in the image the position of these points. Program the Tsai algorithm to recover the camera model. Compare with the known value. Add radial distortion in the image, recompute the camera model and compare. Add different levels of noise in the data, recompute the camera model and compare. (see: section A of Song, Wu and Li, " Camera Calibration Survey"(2014).
  2. Move the camera to a different position and take another picture. Use both views to determine the epipolar line correspondent to a point marked by the user in the other image. Visually compare the result.
  3. Using a feature detection algorithm, find a set of corresponding points, one in each image. Visually compare the result. (see: section A of Humenberger et all, " A fast stereo matching algorithm suitable for embedded real-time systems"(2010).
  4. Find the 3D position of these points in the scene. Compare.
  5. Triangulate the scene and reconstruct a 3D. Compare.
  6. Reconstruct the ZBuffer of the scene. Compare.
  7. Select eight pairs of points, one in each image, and compute the Fundamental Matrix. Compare with know values.

Alternative sugestion: A challange with a soccer match

Consider a video segment of a soccer match. What are the intrinsic and extrinsic parameters of the camera? What is the movement of each player?