Anomaly detection for video surveillance in crowded environments

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Enformatik Enstitüsü, Modelleme ve Simülasyon Anabilim Dalı, Türkiye

Tezin Onay Tarihi: 2014

Öğrenci: CİHAN ÖNGÜN

Eş Danışman: İLKAY ULUSOY, ALPTEKİN TEMİZEL

Özet:

Crowd behavior analysis and anomaly detection in crowded environments have become more important in recent years. In the literature there are two main approaches for crowd behavior analysis based on the density of the crowd. While individual analysis is more efficient for low and medium density crowds, holistic approaches which consider the crowd as a whole are more efficient for high density crowds. Crowd behavior analysis studies can be examined in 3 categories: group behavior analysis, crowd behavior analysis and anomaly detection. While group behavior analysis is based on detection and tracking of human groups, crowd behavior analysis studies considered the whole crowd in the video. These steps are generally followed by anomaly detection which is the task of detecting the events which are normally not expected in a scene. In this work, the aim is to detect behavioral anomalies in high density crowds where detection and tracking of individuals are difficult. Video scene is considered as a whole and a heat map is generated using Finite-Time Lyapunov Exponents (FTLE) based on motion changes and this heat map is divided into behavioral clusters using hierarchical clustering. Then considering the distribution of these clusters existence of anomaly is determined and abnormal cluster are detected using an adaptive threshold.