Computer Vision and Multimedia Analysis - Master Course in Computer Science @ UniTn 2017/18
Francesco Alzetta - francesco.alzetta@studenti.unitn.it
git clone https://github.com/ElDivinCodino/CVPedestrianCounting.gitIn this project I have tried several algorithms in order to do a satisfying traking of cyclists and pedestrians. In the end I found that it was a job hard to achieve without background subtraction. For this reason I decided to mix a performing background subtraction algorithm, as MOG2, set with a high threshold in order to delete the biggest noise, and an algorithm able to recognize pretty well the elements from the remaining background, found at https://github.com/andrewssobral/bgslibrary.
Unfortunately during the tuning of the parameters I understood that this solution is very sensitive, and changing slightly a relevant variable could make the code recognize better the pedestrians but introducing also lot of noise at the same time.
For this reason I decided to loose the detection of the most problematic pedestrians, but avoiding as much as possible false positives.
In the root folder there is a Makefile, where all the dependencies are included.
- Move into the Assignment directory.
- Place the video in the Assignment/res folder (removed for file size purposes).
- In order to compile the code just run the
makecommand from terminal. - To execute the code run
./main.
A video showing how the code performs can be found at https://drive.google.com/file/d/18sfNGjN3pMVY1UqovKRmif4TMGAHA9ha/view?usp=sharing