Vision driven visitor behaviour analysis and crowd management


Overcrowding and queues are frequently reported amongst the most negative experience while visiting large exhibitions. Proper crowd control is thus paramount to an enjoyable visiting experience and is essential to improve the overall fruition of heritage and art. Current crowd management systems mainly rely on human instructors to direct visitors on pacing, gathering and pausing. As such, they potentially lack adequate and intelligent sensing of visitors' behaviours and feelings. An ideal system should predict visitors' attention and discretely guide visitors to freely enjoy exhibits without interruptions while keeping the crowd in order and safety.

We believe the key techniques behind such systems should be high-performance human counting and tracking, activity recognition, expression recognition, automatic visitor analysis, and automatic crowd management. Current public facilities, like museums, are densely equipped with CCTV cameras, yielding vision information in principle readily available for quantitative crowd monitoring and control. In addition to ordinary cameras, cheap and accurate 3D depth sensors can be locally deployed for greatest accuracy on visitor positions, gestures and expression tracking. However, to analyse the collected information and to better design visitors' experience, several aspects associated to the physics of crowds dynamics need to be further understood.

Our objective is to develop cutting-edge computer vision techniques for visitor behaviour analysis and crowd management. This project will combine the expertise of two research groups on image analysis and 3d depth sensors for individual tracking, activity analysis and expression recognition. We will employ deep learning techniques, statistical data analysis and analytical modelling in order to advance the fundamental understanding on crowd dynamics, to develop tools to monitor crowd moods, and to devise automated crowd managing systems for large exhibitions. We will closely collaborate with key players in the fields, such as Philips and Huawei, in order to deploy our system to real-life applications.


Project number


Main applicant

Prof. dr. F. Toschi

Affiliated with

Technische Universiteit Eindhoven, Faculteit Wiskunde en Informatica, Centre for Analysis, Scientific computing and Applications (CASA)