If you have any problems related to the accessibility of any content (or if you want to request that a specific publication be accessible), please contact us at scholarworks@unr.edu.
Robust Event Detection and Retrieval in Surveillance Video
Date
2014Type
ThesisDepartment
Computer Science and Engineering
Degree Level
Master's Degree
Abstract
We developed a robust event detection and retrieval system for surveillance video. The proposed system offers vision-based capabilities for the detection and tracking of various objects of interest, and can recognize events such as: 1. a person with certain attributes being present in the scene; 2. two people meeting; 3. people carrying bags; 4. bags being dropped; 5. bags being stolen; 6. bags being exchanged; 7. two people handshaking; 8. one person's pointing gesture. We use an improved adaptive Gaussian mixture model for background modeling and foreground detection; a connected component labeling algorithm is then employed to label the foreground pixels. A Kalman filter approach is used to build models for the entities of interest (people and bags), which is combined with color histograms for tracking. We use shape symmetry analysis and color histograms to detect people carrying bags. Our experiments demonstrate the ability to search for instances of events according to specific attributes in large video sequences.
Permanent link
http://hdl.handle.net/11714/2875Additional Information
Committee Member | Nicolescu, Monica; Pinsky, Mark |
---|---|
Rights | In Copyright(All Rights Reserved) |
Rights Holder | Author(s) |