Institutional Repository [SANDBOX]
Technical University of Crete
EN  |  EL

Search

Browse

My Space

Detection and semantic analysis of objects and events through visual cues

Makantasis Konstantinos

Full record


URI: http://purl.tuc.gr/dl/dias/7B1432ED-164C-428C-B423-72C0D594D352
Year 2016
Type of Item Doctoral Dissertation
License
Details
Bibliographic Citation Konstantinos Makantasis, "Detection and semantic analysis of objects and events through visual cues", Doctoral Dissertation, School of Production Engineering and Management, Technical University of Crete, Chania, Greece, 2016 https://doi.org/10.26233/heallink.tuc.65351
Appears in Collections

Summary

This dissertation counts in total ten technical chapters, plus a conclusion chapter. Technical chapters are organized in three parts; each one of them is dedicated to a different aspect of semantic visual content analysis. The first part consists of four chapters and entitled ”From Objects to Events”. As the title suggests, in this part, we investigate how the information about objects in a scene can be available as a basis for event understanding. There are three different technical chapters, in which we try to address three different real-world problems; the development of (i) a supportive vision based system for detecting in real-time elderly and/or patients fall in indoor environments, (ii) a maritime security vision based system and (iii) a surveillance system for activity recognition in industrial workflow.The second part entitled ”From Unstructured Visual Content to Objects”. In this part we investigate how the visual content that is stored in distributed and heterogeneous Internet databases can be, initially, organized, and then utilized towards objects documentation. Specifically, in this part we propose (i) a method for retrieving and dynamically clustering user generated photographs available over the web and (ii) and online image indexing scheme.The third and last part entitled ”Beyond the Visual Spectrum” and focuses on visual content analysis using thermal and hyperspectral data. There are two technical chapters; the first one presents an algorithm for background subtraction applied on thermal video streams, while the second one presents a method for material recognition using hyperspectral images.We approach each one of the aforementioned problems through the levels of understanding framework. Initially, we formulate in detail the problem at hand along with its constraints and specifications, explaining what computations will do and why they will do it. Then, we proceed with proposed solution design and implementation, where we describe in detail the tools for developing the proposed solutions, the input and output of the system as well as, all intermediate representations of visual information. Finally, we evaluate proposed solutions performance on both synthetic and real-world data.

Available Files

Services

Statistics