Your browser does not support JavaScript!

Home    Search  

Results - Details

Search command : Author="Στεφανίδης"  And Author="Κωνσταντίνος"

Current Record: 20 of 47

Back to Results Previous page
Next page
Add to Basket
[Add to Basket]
Identifier 000392818
Title Tracking, re-identification and interaction using RGB-D sensors in ambient intelligence environments
Alternative Title Παρακολούθηση, επαναταύτιση και αλληλεπίδραση με χρήση αισθητήρων RGB-D σε περιβάλλοντα διάχυτης νοημοσύνης
Author Γαλανάκης, Γεώργιος Ε.
Thesis advisor Στεφανίδης, Κωνσταντίνος
Reviewer Αργυρός, Αντώνιος
Ζαμπούλης, Ξενοφών
Abstract In recent years, the emergence of RGB-D sensors has reinforced the formulation of Computer Vision approaches which support natural interaction in unobtrusive ways. Such types of interaction are useful in Ambient Intelligence environments, whose purpose is to either entirely hide the presence of technology from users or to smoothly integrate technology within the environment. In this context, this thesis addresses the problems of person tracking, person re-identification and recognition of head gestures. In this thesis, the person tracker is a system which analyzes video input to detect and monitor the positions of multiple persons over time. The tracker uses multiple cameras as they facilitate coverage of wider areas; moreover, a sophisticated method to cope with occlusions is not required. Common challenges appear when multiple persons are in the same place, come close or even interact with each other. The proposed multi-person tracking system receives input from multiple RGB-D sensors which share a common field of view, constructs a 3D representation of the scene and, then, detects and tracks persons. To address possible ambiguities during the assignment of tracks, the system combines position prediction and appearance characteristics. A per son's appearance is projected in a 3D representation that captures the spatial arrangement of colors. The proposed system is evaluated using complex datasets and compared against two former tracking approaches. The coverage of the person tracker is limited to a single scene, thus an expansion of its functionality is the re-identification of a person who visits multiple scenes. This thesis presents an architecture that combines information from multiple person trackers; each one monitors a distinct scene. Its main component is a re-identification system which maintains a database. This database consists of the 3D representations of persons which are maintained during the tracking. As soon as a person enters the scene, techniques based on appearance characteristics are applied in order to match with a database entry. To measure the accuracy of the system, an evaluation based on offline data was performed. Reported results indicate adequate accuracy, which however decreases when the appearance representations are obtained under significantly different lighting conditions. This thesis also presents a system which recognizes head gestures by analyzing a person's head motion. To this purpose, a head pose estimator, based on depth data, is utilized. To characterize the head motion, preliminary experiments were conducted, indicating properties and anatomical limits of this motion. The proposed approach detects a set of primitive gestures. The method is evaluated within the context of a human-computer dialog, yielding competitive recognition results to state-of-the-art approaches. The evaluation to the aforementioned systems indicated successful adoption of RGB-D sensors to address the pursued problems of person tracking, person re-identification and recognition of head gestures. Discussion upon future work concerns the integration of the above approaches in everyday environments.
Language English, Greek
Subject Διάχυτη νοημοσύνη
Issue date 2015-07-17
Collection   Faculty/Department--Faculty of Sciences and Engineering--Department of Computer Science--Post-graduate theses
  Type of Work--Post-graduate theses
Permanent Link https://elocus.lib.uoc.gr//dlib/f/f/0/metadata-dlib-1431330747-471368-9431.tkl Bookmark and Share
Views 420

Digital Documents
No preview available

View document
Views : 43