DSpace Repository

Human behavior profiling for a video surveillance system

Show simple item record

dc.contributor.advisor Dailey, Matthew N. (Chairperson) en_US
dc.contributor.author Kan Ouivirach en_US
dc.date.accessioned 2015-01-12T10:40:56Z
dc.date.available 2015-01-12T10:40:56Z
dc.date.issued 2008-05 en_US
dc.identifier.other AIT Thesis no.CS-08-01 en_US
dc.identifier.uri http://www.cs.ait.ac.th/xmlui/handle/123456789/359
dc.description 48 p. en_US
dc.description.abstract Due to increasing crime, video surveillance systems are being deployed in more and more places. Video surveillance systems are needed to help security personnel prevent and respond to criminal activity in time. However, most systems have many limitations. Manual review of all video footage is too time-intesive and error-prove. Motion detection helps reduce the amount of data need in manual review, but it is useless in complex scenarios in which there is always motion. Therefore, in this thesis, I explore a general-purpose approach to human behavior pro ling and anomaly detec- tion using Hidden Markov Models (HMMs). I extract features from a moving object for describing human behavior. Then I compare the use of discrete-density HMMs using vector quantization via the k-means algorithm and continuous-density HMMs using single Gaussian distributions to evaluate which model is best for anomaly detection. From my experiments, the results demonstrate that both models correctly recognize predi ned normal human behaviors and identify anomalous behavior in a scene; how- ever, discrete models preform better than continuous models for prede ned normal behavior. Unfortunately, discrete-density HMMs with vector quantization are di cult to evaluate in online learning. Further research is required on using continuous-density HMMs as discriminative models in online learning scenarios. en_US
dc.description.sponsorship Royal Thai Government en_US
dc.language.iso en en_US
dc.publisher Asian Institute of Technology en_US
dc.relation.ispartofseries AIT Publications; en_US
dc.subject Fetal monitoring en_US
dc.subject Human behavior -- Mathematical models en_US
dc.subject Electronic surveillance en_US
dc.title Human behavior profiling for a video surveillance system en_US
dc.type Thesis en_US

Files in this item

This item appears in the following Collection(s)

Show simple item record

Search DSpace

Advanced Search


My Account