National and Kapodistrian University of Athens
Εξαγωγή σημασιολογικής γνώσης από γεγονότα
Abstract
dc:descriptionThis work reports on research conducted in the domain of semantic information extraction from events in video streams. The research focuses on exploring techniques used for human behavior recognition from video sequences and creating complete systems for anomaly detection observed in the video, human behavior recognition and video indexing. Behavior understanding from events can be considered as a typical classification problem using predefined classes. In our research presented in this thesis, the main idea is to deal with this classification problem from a bottom up point of view using motion analysis for behavior recognition. We describe our research work from its first steps to the final accomplishments including the development of surveillance systems which can to operate in off-line or real time conditions. Following a theoretical analysis of the problem (including definitions of short term behaviors, events and long term behaviors) we follow preprocessing steps and calculate features that form the input of a higher level behavior recognition module. Novel methods are introduced for behavior classification and semantic representation from events. A methodology is proposed whereby frame information is coded in graphs, using only optical flow to form the feature vector aiming in behavior recognition. The proposed graphs (called Optical Flow Proximity Graphs - OFPGs) are used in order to hold information about neighboring values of the optical flow. Operations on graphs transform the initial problem of representation to a classification problem. A symbolic method including two levels of graph representation is also proposed dealing with the same recognition problem. OFPGs are also proposed for video indexing. Furthermore, a bottom-up approach for human behavior understanding is presented, using a multi camera system for anomaly detection. In the framework of this system, we present the use of one class continuous Hidden Markov Models (cHMMs) for the task of human behavior recognition. An approximation algorithm, referring to the Forward Backward procedure of the cHMM, is proposed to overcome numerical stability problems in the calculation of probability of emission for very long observations, called Observation Log Probability Approximation (OLPA). Experimental results using open source data sets, such as PETS04 and Weizmann, are compared with results of other state of the art methods, illustrating the advantages of the proposed methods.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2005
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Αντωνακάκη, Παναγιώτα
- Antonakaki, Panagiota
Subjects
dc:subject × 12Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/29136