Crowd counting applied in UCLouvain auditoriums

Details

Supervisors

Schaus, Pierre

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en informatique, à finalité spécialisée

Abstract

With the growing number of people in the world, crowd counting has become more and more useful over the years. Until recently, it was done manually by human people with a poor precision and very different numbers. The recent advances in the domain of machine learning now permits to create performing crowd counting algorithms based on images. Since they installed security cameras in most auditoriums, the University of Louvain-la-neuve can also benefit of crowd counting analysis to improve the allocations of the auditoriums and other resources. The main goal of this thesis is to try to improve the accuracy of the predictions by training already existing models on the auditoriums. We trained and analysed different models that had been trained in different ways. This allowed us to determine which models are the most efficient and which are the most versatile. Moreover, in order to make the work we have done more accessible, we have created an interface and a python library which we hope will one day be useful.