Project Details

Host Institution:Democritus University of Thrace
Project Duration: Mar. 2012 - Mar. 2015
Principal Investigator: Nikolaos Mitianoudis, DUTH, GR
Post-Doctoral Researcher: Dr. Dimitrios Alexiadis
Foreign University Liaison: Dr. Tania Stathaki, Imperial College London, UK.
Funding Source: General Secretariat of Research and Technology, Greece. Action 'Support of post-doctoral researchers'
Total Funding: 96.000E

Project Objectives

Images are produced as projections of the 3D world on the 2D plane of light-sensitive sensors (image plane) of a capturing device. The 3D motion of an object results into 2D motion of its projection on the image plane. The 2D motion of each image point is referred to as the “2D motion field”, also known as the “optical flow”. Although the problem of 2D motion estimation has been extensively studied in the literature, it is still an active research topic, due to its importance in many challenging applications including video coding, passive scene interpretation, object tracking, visual surveillance systems, content-based video retrieval and robotic vision.

The proposed project is aiming to extend current research in frequency-domain approaches for 2D motion estimation to the problem of estimating the actual 3D motion and structure from sequences of 3D representations over time. This problem has not been studied yet in the frequency domain. During the project, we will extend the derived frequency-domain techniques towards two important and challenging application fields, where motion estimation is required: a) Estimation of 3D motion in sequences of 3D (volumetric) data, b) Joint estimation of 3D structure and motion from sequences of binocular or multi-view color images.

Project Outline

F3SME is divided into the following work packages and Deliverables.

Component 1

Novel algorithms for 3D motion estimation from volumetric data sequences


Activity no. Description
1.1 Bibliographic research on 3D motion estimation algorithms from volumetric data
1.2 Collection of volumetric test data – Capturing setup and creation of appropriate dataset
1.3 Mathematical analysis of the 3D motion estimation problem in the frequency domain
1.4 Construction of multidimensional directional filters, appropriate for 3D motion estimation
1.5 Initial formulation of frequency-domain and/or filter-based 3D motion estimation algorithms
1.6 Testing and refinement of the algorithms
1.7 Implementation of the algorithms in CUDA
1.8 Dissemination activities

Component 2

Novel algorithms for structure and motion estimation from multi-view image sequences


Activity no. Description
2.1 Bibliographic research on structure and motion estimation from multi-view image sequences
2.2 Collection of test multi-view videos – Capturing setup and creation of appropriate dataset
2.3 Mathematical analysis of the joint structure and motion estimation problem in the frequency domain
2.4 Construction of multidimensional directional filters, appropriate for structure and motion estimation from multi-view video
2.5 Initial formulation of frequency-domain and/or filter-based algorithms
2.6 Testing and refinement of the algorithms
2.7 Development of a complete C++ library
2.8 Dissemination activities