Font Size: a A A

Research On Pedestrian Behaviors In Video Sequences

Posted on:2009-04-15Degree:MasterType:Thesis
Country:ChinaCandidate:X K WuFull Text:PDF
GTID:2178360278956633Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
The widespread application of video made people's social life more wonderful, but along with it the massive video data put forward higher demand to video processing technology. So the research on improving the automation and intelligence of video processing is paid more and more attention inevitably. And as primary organizer of many social activities, human is regard as an important concerned object of many video data naturedly. Hence the vision analysis of human behavior becomes one of the advanced research directions in the field of video processing and computer vision in recent years.This thesis focuses on analysis of pedestrian behaviors in video. Our main work and research results involve the following aspects: proposes a new processing flow; researches into methods of foreground video clips detection and classification; and developes modeling processing of pedestrian behaviors analysis.A new method to analysis pedestrian behaviors in video is presented, which is different from traditional processing flows. It includes three steps: (1) the foreground video clips in a video sequence are detected, where each foreground video clip contains some kind of pedestrian behaviors; (2) then the foreground video clips in a video sequence are classified; (3) the pedestrian behaviors in each foreground video clip classified are described and interpretated.New approaches are developed in foreground video clips detection and classification respectively. Two detection approaches are presented: one is based on histogram distance and other is based on PCA. There is one approach to foreground video clips classification, where blob of object and height-width ratio of its minimum enclosing rectangle are introduced as new characters, then histogram statistics is applied to height-width ratio sequences in clips classification.HMM is presented to describe pedestrian behaviors in clips, whose features are height-width ratio sequences, and based on HMM, interpretations of the pedestrian behaviors are carried out.
Keywords/Search Tags:PCA, HMM, Foreground Video Clips Detection and Classification, Behavior Analysis
PDF Full Text Request
Related items