Snelle persoonsheridentificatie op grote schaal
In this PhD, the topic of person re-identification is studied for real-life industrial applications.
Recent techniques realize the generation of a robust descriptor for a pedestrian, such that
several appearances of the same person in different situations (camera viewpoint, moment in
time) can be matched in a way that the person’s identity can be coupled between these two
views. However, several challenges remain in this field that form a hindrance for applying
these techniques in real-life industrial situations. Example applications are the monitoring of
customer behavior in large shopping malls, the search for certain individuals in the
surveillance camera footage of an entire city and the construction of an individual photo
album of a certain child from the pictures that are taken during an entire school year. The
main challenge in these applications is obviously the large amount of data through which
such a person re-identification must search. Moreover, in many applications the processing of
multiple camera streams must run in real-time, preferably on hardware that is not too costly.
Application and site-specific know-how must evidently be exploited to cope with these