SPATMODEL - Spatial Constellation Models of Local Object Descriptors
The main goal of this project is to study existing and develop new methods for detecting, localising and recognising objects based on extracted local image features. The methods should integrate both the local feature information and their spatial configuration (constellation) in order to find "best hypotheses" of objects and their pose. The fundamental idea is to utilize a "constellation model! which tolerates missing evidences and is efficient in search.
This project is a sub-project of the more general project: OBJECT - Object Detection and Recognition, and the results will be utilized in the main project.The current main objectives are:
- State-of-the-art method for finding objects based on the extracted local image features
- RANSAC based hypothesis initialisation and MCMC based object presence and pose estimation
The first very impressive results will be demonstrated in ICCV2007 workshop on Non-rigid Registration and Tracking Through Learning!
|Alexander Drobchenko||WWW||Post-graduate researcher|
|Joni Kämäräinen||WWW||Project leader, lecturer|
|Research groups working on constellation models|
|CalTech Vision Group (prof Perona)|
|Univ. British Columbia - Lab for Computational Intelligence (prof Lowe)|
|Related or otherwise useful software|
|Drop - Deformable Regisration using Discrete Optimisation|
3D pose estimation (i.e. camera resectioning)
- The Resultant and Bezout's Theorem
- This can be used to solve common roots of lower order polynomials by using matrix determinant of higher order polynomials (used by at least one method published in PAMI)
Do not hesitate to to contact authors (e-mail, etc.) in order to retrieve copies or reprints of the following publications.