École Nationale Supérieure de
Techniques Avancées

Abstract

This internship deals with people detection and tracking on the Domplatz, a square of Hamburg, with HD cameras. It is all about deciding which method is the best for the project, taking into account the constraints of the situation. The work is divided in three parts : first, a survey of the state-of-the-art methods, then the creation of a decision tool for these methods and less advanced methods. Finally, we made a benchmark of the most relevant algorithms, to decide in the end which algorithm suits best the following project. The detectors have to be trained on the Caltech Pedestrian Dataset. The most promising one is the Faster RCNN. The tracking are performing well enough to be used in the rest of the project.