Computer Vision Mini-projects:
The mini-project: The mini-project consists on your own implementation of a algorithm or a combination
of algorithms to solve a particular problem in computer vision. It can also be a modification of an existing
implementation (open source). The (modified) source code also needs to be submitted to Stream.
Preferably the mini-project should be developed with groups of two students.
You should prepare a small report, similar to a conference paper (between 8 to 10 pages using the given
template). Use the report template (either Word or Latex) available on Stream.
The report must cover the following sections:
– Abstract (short description of the project, summary of the results)
– Introduction or Background (a mini-literature review): a discussion about a few articles/books that
presents the related topics (remember to cite your sources appropriately).
– Methodology: describe, with your own words, the methods and/or algorithms presented.
– Results: show results you obtained and compare to the results from the literature.
– Discussion: advantages and limitations, comparisons with other methods etc.
– Conclusions: your conclusions regarding the effectiveness of the method, what has been learnt, future
– Bibliography: list the relevant articles and/or books you have used. Use a citation style (see the
template instructions on StreamTopics
Methods and/or applications for computer vision that we did not discuss in details during our classes are
topic candidates. The key is not to implement everything from scratch, but to experiment with different
methods, compare two or more methods or explore a different way of using a certain method.
After choosing a topic, please approach the lecturer to formalise your choice and to confirm that there is
no other group with the same topic.
Brief description of the topics available in 2021:
Depth maps with stereo vision: Measure and compare the real distances with that of the stereo vision
algorithms. Change the distance between the two cameras to see how to improve the accuracy (and by
High dynamic range approach: Implement a simple colour classifier based on multiple cameras based on
HDR. Compare the results using one camera against more than one.
Traffic sign recognition: Use a well known database to train your own approach to recognise traffic signs.
Compare the results.