Anomaly detection using sparse features and spatio-temporal hidden markov model for pedestrian zone video surveillance

Download
2014
Gündüz, Ayşe Elvan
Automated analysis of crowd behavior for anomaly detection has become an important issue to ensure the safety and security of the public spaces. Public spaces have varying people density and as such, algorithms are required to work robustly in low to high density crowds. Mainly, there are two different approaches for analyzing the crowd behavior: methods based on object tracking where individuals in a crowd are tracked and holistic methods where the crowd is analyzed as a whole. In this work, the aim is to detect anomalies in pedestrian zone videos using a holistic approach. The pedestrian zone videos are automatically grouped according to crowd density. The pedestrian motion is modeled as a whole without detecting and tracking the individuals using the features obtained Oriented Fast and Rotated Brief (ORB) feature detector and thus the model is privacy preserving. These features are then represented using Binary Robust Independent Elementary Features (BRIEF) descriptor and a spatiotemporal Hidden Markov Model is used for anomaly detection.
Citation Formats
A. E. Gündüz, “Anomaly detection using sparse features and spatio-temporal hidden markov model for pedestrian zone video surveillance,” M.S. - Master of Science, Middle East Technical University, 2014.