Crowd Detection and Modeling for People Counting and Motion Analysis

  • Mohammed Sami Zitouni

Student thesis: Master's Thesis


Visual recognition of crowd dynamics has had a huge impact on several applications including surveillance, situation awareness,homeland security and intelligent environments. However,the state-of-the-art in crowd analysis has become diverse due to factors such as: a) the underlying dentition of a crowd, b) the constituent elements of the crowd processing model, c) its application, hence d) the dataset and e) the evaluation criteria available for benchmarking. Although such diversity is healthy, the techniques for crowd modeling thus developed have failed to establish credibility therefore becoming unreliable and of questionable validity across different research disciplines. The objective of this research work is to develop a framework for crowd detection and modeling and apply it in people counting, people tracking, and crowd motion analysis applications. The thesis proposes a novel motion-based crowd detection technique using a combination of Gaussian Mixture Models (GMM) and Dynamic Textures (DT). A spatio-temporal and iterative implementation of the method is introduced to enable the technique to handle dynamic scenes represented in the form of moving backgrounds. Further, a hierarchical crowd detection and representation framework that combines motion detection and appearance saliency is also proposed. This framework allows crowd analysis to be performed at either or both in the macro (holistic) or micro (individual) levels independent of the crowd density. Finally, two application case studies demonstrating the use of the proposed hierarchical crowd model in people counting and crowd motion analysis are considered. In the people counting study, the proposed framework is shown to operate between macro and micro levels where crowd density estimation is obtained by individuals detection and their agglomeration in a hierarchical manner. In addition, in the crowd motion analysis application, macro level tracking is based on the state estimation of the crowd which is inherited from the GMM of DT, and micro level tracking of individual people is facilitated using Kalman ltering. The research proposed in this thesis has explored the possibility of modeling crowd at both macro and micro levels through the use of a hierarchical crowd detection and representation framework. The framework, in addition to being accurate also provides the exibility to analyze crowd information at various levels of semantic abstraction thus providing future access to critical situation awareness applications. Indexing Terms: Crowd Modeling, Motion Detection, Appearance Saliency, People Counting, Tracking and Motion Analysis.
Date of AwardDec 2015
Original languageAmerican English
SupervisorHarish Bhaskar (Supervisor)


  • Crowd Modeling
  • Motion Detection
  • Appearance Saliency
  • People Counting
  • Tracking and Motion Analysis.

Cite this