Matematikcentrum

Lund University

  • Titel: Adherent raindrop modeling and detection
  • Kort beskrivning:

    Develop and test existing algorithms for detecting raindrops on the frontglas/dome on Axis cameras. The algorithm should be accurate, efficient and can not be CPU intensive since it should smoothly run on a 1 GHz CPU camera

  • Lång beskrivning:

    Develop and test existing algorithms for detecting raindrops on the frontglas/dome on Axis cameras.
    The algorithm should be accurate, efficient and can not be CPU intensive since it should smoothly
    run on a 1 GHz CPU camera.
    The outline of the project is to first do a short literary study to see what algorithms that might
    useful. A good starting point for the students is the paper by You et al. [1] on adherent raindrop
    modeling and detection where they exploit local spatio-temporal derivatives of raindrops. There are
    also other methods proposed for adherent raindrop detection [2, 3, 4, 5], but models for falling
    raindrops might also be useful [6, 7, 8] in combination. Once a few different models have been
    identified, own ideas are also welcome (for simple detection recipe see [9]), the students will
    translate these methods into simple code. They then get a few cameras with a dome and front glass
    to test the different methods on. This is achieved by running the algorithms on host computer
    connected to the cameras. The tests are ideally done both in the lab and in real outdoor scenes. The
    students will finally evaluate the tests and identify the most promising method. If there still is time,
    the code should be translated into C code and integrated into Axis cameras.

  • Info: PDF
  • Kontaktperson: Anders Heyden